MR Y 38254 PDF

AC Microgrids Analysis, Optimization and Planning for
Resilience Enhancement
Thèse
Youthanalack Vilaisarn
Doctorat en génie électrique

Philosophiæ doctor (Ph. D.)
Québec, Canada
© Youthanalack Vilaisarn, 2022

AC Microgrids Analysis, Optimization and Planning for
Resilience Enhancement
Thèse
Youthanalack Vilaisarn
Sous la direction de :
Jérôme Cros, directeur de recherche

Morad Abdelaziz, codirecteur de recherche
Résumé
Face à des événements météorologiques violents, un système de distribution électrique peut souffrir
de la perte ou de la défaillance d'un ou de plusieurs de ses composants. Ce phénomène est connu sous
le nom de contingence de système. Néanmoins, en tirant parti des systèmes de protection, de
l’électronique de puissance et de la pénétration des ressources énergétiques décentralisées dans le
réseau électrique, un système de distribution électrique a la possibilité d'être reconfiguré en micro-
réseaux. Cela permet de résister contre de telles éventualités en gardant au minimum la possibilité
d’une interruption de l'alimentation. Poussé par des facteurs techniques, économiques et
environnementaux, ainsi que par le déploiement rapide d'un grand nombre de ressources de
production décentralisées, le micro-réseau est récemment devenu un concept important dans un
système de distribution actif et rapidement reconfigurable.
Les micro-réseaux ont la capacité de fonctionner à la fois en mode connecté au réseau et en mode
isolé. En raison de cet avantage, le micro-réseau est devenu un élément clé du futur réseau intelligent.
Bien que le concept de micro-réseau puisse apporter différents avantages aux services de distribution
et aux clients, à savoir une amélioration de l'économie, de l'environnement et de la résilience, il offre
toujours un défi au niveau de la planification et la gestion opérationnelle. Les défis de la planification
des micro-réseaux proviennent de : 1) la nature intermittente et incertaine des ressources
décentralisées et des charges des systèmes distribués, ainsi que l'incertitude relative aux contingences
auxquelles le réseau de distribution est confronté, 2) la charge de calcul que la prise en compte des
incertitudes du micro-réseau implique, et 3) le grand nombre de compromis entre les différents
objectifs d'optimisation possibles du micro-réseau qui doivent être pris en compte dans la phase de
planification. Motivée par ces défis, cette recherche propose le développement de nouvelles
méthodologies d'analyse et de planification qui peuvent assurer l'efficacité du processus de création
du micro-réseau en tenant compte des caractéristiques particulières et de la philosophie opérationnelle
du micro-réseau.
Dans un premier temps, les modèles d’étude d’écoulement de puissance linéaires et non linéaires sont
développés en prenant compte des caractéristiques réelles d'un micro-réseau insulaire équilibré et
déséquilibré, c'est-à-dire l'absence de bus infini, la présence d’une fréquence du système variable et
de certains générateurs fonctionnant en mode de contrôle « statisme ». Tout d'abord, nous présentons
un algorithme non linéaire basé sur la méthode des mailles et la matrice Zbus pour un micro-réseau
opéré en mode « droop ». Cet algorithme sans inversion est particulièrement adapté aux grandes
dimensions des systèmes de distribution pratiques comprenant des milliers de nœuds électriques,
ii
lorsqu'ils fonctionnent comme des micro-réseaux insulaires. Deuxièmement, un modèle linéaire pour
étudier l’écoulement de puissance (LPF) basé sur la méthode des nœuds est proposé. Étant basé sur
la méthode des nœuds, le modèle proposé est utilisable dans différents problèmes d'optimisation de
micro-réseaux. Des études de cas numériques sont développées et utilisées pour démontrer la
précision des modèles proposés et garantir leur application réussie pour modéliser avec précision
l’écoulement d'énergie dans un micro-réseau. Cela permet son application dans un processus
d’optimisation présenté à l’étape suivante de cette recherche.
Dans la deuxième étape, cette recherche propose un modèle d’écoulement de puissance optimal (OPF)
pour le fonctionnement optimal des micro-réseaux à courant alternatif, équilibrés et déséquilibrés,
avec un contrôle hiérarchique (c'est-à-dire, un contrôle primaire et secondaire). Le modèle proposé a
d'abord été formulé comme un modèle non linéaire en nombres entiers (MINLP), puis il a été linéarisé
et converti en un modèle linéaire en nombres entiers (MILP) en utilisant le modèle LPF développé
dans la première étape de cette recherche. La philosophie de fonctionnement du micro-réseau, en
mode connecté au réseau et en mode îloté, a été prise en compte. De plus, plusieurs types de
générateurs distribués, y compris ceux qui sont disptachables et non disptachables, ainsi que des
ressources de stockage d'énergie, ont été pris en compte dans le modèle MILP proposé. Plusieurs
études de cas numériques ont été menées pour valider et prouver l'efficacité et la précision du modèle
MILP développé. Les résultats de ces études de cas ont démontré la précision et la supériorité de
calcul du modèle MILP proposé.
Enfin, un cadre de planification pour les micro-réseaux dans les réseaux de distribution actifs est
proposé. Ce cadre de planification vise à améliorer la résilience des systèmes de distribution
d'électricité face à des événements de faible probabilité, à fort impact. Dans le cadre proposé, le
problème de planification a été présenté comme un problème d'optimisation stochastique à deux
niveaux. Tout d'abord, le niveau externe traite du placement optimal des éléments de planification du
système de distribution (c.-à-d. les ressources répartissables ou non répartissables, les unités de
stockage d'énergie et les interrupteurs d'isolement). Ce problème a été formulé en utilisant une
formulation d'optimisation multi-objectifs et ensuite l'algorithme métaheuristique bien connu NSGA-
II est adopté pour la recherche d’une solution optimale. Cette approche permet de déterminer les
solutions qui impliquent le meilleur compromis entre plusieurs objectifs éventuellement conflictuels
du problème de planification, à savoir le coût, la résilience et l'impact environnemental.
Deuxièmement, le niveau interne du cadre de planification traite du problème d'optimisation relatif
au fonctionnement optimal des micro-réseaux qui peuvent être créés par les éléments de planification
du système de distribution alloués dans le niveau externe. Le problème du fonctionnement optimal
iii
du micro-réseau est présenté comme un problème d’étude de l’écoulement de puissance optimal
linéaire (LOPF). À cette fin, le modèle MILP développé dans la deuxième étape est adopté.
Néanmoins, il est nécessaire de prendre en compte différents scénarios stochastiques dans le niveau
interne pour tenir compte des différentes incertitudes du système. Il faut aussi considérer la nature
métaheuristique du niveau externe ce qui demande la résolution du modèle LOPF pour chacun des
scénarios stochastiques et aussi pour chaque individu de la population. La prise en compte de ces
facteurs présente un défi au niveau du calcul. Par conséquent, une nouvelle méthodologie utilisant un
modèle de réseau de neurones (DNN) est proposée. Cette méthode permet de dériver rapidement
l'information requise des solutions LOPF pour les scénarios stochastiques considérés. Enfin,
l'efficacité du cadre proposé est validée par des résultats de simulation numérique.
iv
Abstract
Facing severe weather events, a distribution system may suffer from the loss or failure of one or more
of its components, known as N-K contingencies. Nevertheless, taking advantage of the system’s
isolate switches and the penetration of the distributed energy resources in the electrical grid, a
distribution system has the possibility to be clustered into microgrids in order to withstand such
contingencies with minimal power interruption. Driven by technical, economic and environmental
factors, as well as by the rapid deployment of a large number of distributed generation resources, the
microgrid has recently become an important concept in the active distribution system.
Microgrids have the ability to operate in both grid-connected and islanded modes. The benefits that
the microgrid concept can bring to the operation of the distribution grids make the microgrid a key
component of the future smart grid. While, the microgrid concept can bring different benefits to both
distribution utilities and customers i.e., economic, environmental and resilience enhancement; the
planning and operational management of microgrids still present several challenges for the decision
maker and the distribution network operator. The challenges with the planning of microgrids arise
from: 1) the intermittent and uncertain nature of the distributed energy resources and loads as well as
the uncertainty pertaining to the contingencies facing the distribution network, 2) the computational
burden that considering the microgrid’s uncertainties entails, and 3) the large number of trade-offs
between the different possible microgrid optimization objectives that need to be considered in the
planning stage. Motivated by these challenges, this research proposes the development of new
analysis and planning methodologies that can ensure the efficacy of the microgrid creation process
considering the microgrids special features and operational philosophy.
Initially, nonlinear and linear power flow models are developed to cope with the real characteristics
of balanced and unbalanced islanded microgrid i.e. the absence of the slack bus, the system frequency
being a variable and some DGs operating in droop-control mode. First, a non-linear branch-based
𝐙𝐛𝐮𝐬 algorithm for the droop-controlled islanded microgrid is introduced. This algorithm is inversion
free and is particularly suited for the large dimensions of practical distribution systems comprising
up to thousands of electrical node, i.e., when operated as islanded microgrids. Secondly, a node-based
linear power flow (LPF) model for droop-controlled islanded microgrids is proposed. The node-based
nature of the proposed LPF model, allows this model to be integrated in different microgrid
optimization models. Numerical case studies are developed and are used to demonstrate the accuracy
of the proposed power flow models and guarantee its successful application to accurately model the
microgrid power flow in the optimization application in the next stage of this research.
v
In the second stage, this research proposes an optimal power flow (OPF) model for the optimal
operation of balanced and unbalanced AC microgrids with hierarchical control (i.e., primary droop
and secondary control). The proposed model has been first formulated as a mixed integer nonlinear
programing (MINLP) model, then it was linearized and converted into a mixed integer linear
programing (MILP) model by adapting the LPF model developed in the first stage of this research.
The operating philosophy of the microgrid, in both grid-connected and islanded modes of operation,
was considered. Additionally, several types of distributed generators, including dispatchable and non-
dispatchable, as well as energy storage resources, were considered in the proposed MILP model.
Several numerical case studies were conducted to validate, and prove the effectiveness and the
accuracy of the developed MILP model. The results from the developed case studies demonstrated
the accuracy and the computational superiority of the proposed MILP model.
Finally, a planning framework for microgrids in active distribution networks is proposed. The
proposed planning framework is aimed at enhancing the resilience of power distribution systems
facing high impact low probability events. In the proposed framework, the planning problem has been
casted as a stochastic bi-level optimization problem. First, the outer level deals with the optimal
placement of the distribution system planning elements (i.e., dispatchable/non-dispatchable
resources, energy storage units and isolating switches). This problem has been formulated using a
multi-objective optimization formulation and the well know metaheuristic Non-dominated Sorting
Genetic Algorithm (NSGA-II) algorithm is adopted for its solution. This approach allows for
determining the solutions that entail the best trade-off between the possibly conflicting multi-
objectives of the planning problem, namely, cost, resilience and environmental impact. Second, the
inner level of the planning framework handles the optimization problem pertaining to the optimal
operation of the microgrids that can be created by the distribution system planning elements allocated
in the outer level. The problem of the microgrid’s optimal operation is casted as a Linear Optimal
Power Flow (LOPF) problem. To this end, the proposed MILP model developed in the second stage
is adopted. Despite using a LOPF model, considering different stochastic scenarios in the inner level,
to account for the different system uncertainties, along with the metaheuristic nature of the outer level
make solving the LOPF model for each of the stochastic scenarios for each individual in the
metaheuristic optimization’s population, using a numerical optimization solver computationally
challenging. Motivated by this challenge, a novel methodology using a deep neural network (DNN)
model is proposed for deriving the information required from the LOPF solutions for the stochastic
scenarios under consideration. The effectiveness of the proposed framework is finally validated by
numerical simulation results.
vi
Table of Contents
Résumé ............................................................................................................................................... ii
Abstract .............................................................................................................................................. v
Table of Contents ............................................................................................................................ vii
List of Figures ................................................................................................................................... xi
List of Tables .................................................................................................................................. xiv
Nomenclature.................................................................................................................................. xvi
Remerciements ........................................................................................................................... xxxiv
Foreword ...................................................................................................................................... xxxv
Thesis Original Contribution .................................................................................................... xxxix
Introduction ....................................................................................................................................... 1
0.1 Thesis motivation ...................................................................................................................... 1
0.2 Research Objective.................................................................................................................... 3
0.3 Thesis Outline ........................................................................................................................... 4
Chapter 1 Research Background and State of Art ........................................................................ 6

1.1 Introduction ............................................................................................................................... 6
1.2 Passive and Active Distribution System in the Power System.................................................. 6
1.2.1 Microgrids .......................................................................................................................... 7
1.2.2 Basic Droop Controlled Microgrid and the Implementation.............................................. 9
1.3 Conventional Power Flow and Introduction to Microgrid Power Flow .................................. 11
1.4 Optimum Operation................................................................................................................. 13
1.5 Active Power Distribution System and Microgrid Planning ................................................... 14
1.6 Literature Survey ..................................................................................................................... 16
1.6.1 Islanded Microgrids Power Flow ..................................................................................... 16
1.6.2 Microgrid Optimal Power Flow ....................................................................................... 19
1.6.3 Power System Resilience Enhancement........................................................................... 20
1.6.4 Discussion of the Literature Survey ................................................................................. 23
Chapter 2 Power Flow Algorithm for Islanded MGs .................................................................. 24

Résumé .......................................................................................................................................... 24
Abstract ......................................................................................................................................... 24
2.1 Introduction ............................................................................................................................. 25
2.2 Proposed Algorithm ................................................................................................................ 25
2.3 Extended to Unbalanced Systems ........................................................................................... 29
2.4 Algorithm Validation and Numerical Results ......................................................................... 35
2.4.1 Balanced Microgrids ........................................................................................................ 35
2.4.2 Unbalanced Microgrids .................................................................................................... 37
vii
2.4.3 Microgrids with voltage and frequency dependent loads ................................................. 39
2.5 Conclusion............................................................................................................................... 41
Chapter 3 Linear Power Flow Model for MG system ................................................................. 42

Résumé .......................................................................................................................................... 42
Abstract ......................................................................................................................................... 42
3.1 Introduction ............................................................................................................................. 42
3.2 Formulation and Linear Power Flow ...................................................................................... 43
3.3 Validation of the Algorithm .................................................................................................... 46
3.4 Conclusion............................................................................................................................... 49
Chapter 4 MILP Model for MG’s Optimum Operation ............................................................. 50

Résumé .......................................................................................................................................... 50
Abstract ......................................................................................................................................... 50
4.1 Introduction ............................................................................................................................. 51
4.2 MINLP Problem Formulation ................................................................................................. 52
4.2.1 Objective function ............................................................................................................ 52
4.2.2 Constraints........................................................................................................................ 53
4.2.2.1 Power flow constraints .............................................................................................. 53
4.2.2.2 DDG operation constraints ........................................................................................ 54
4.2.2.3 Energy storage operational constraints...................................................................... 58
4.2.2.4 Power exchange constraints ...................................................................................... 59
4.3 Linearize Formulation ............................................................................................................. 60
4.3.1 Linearization Approach .................................................................................................... 60
4.3.1.1 Linearization of Absolute Value Operator ................................................................ 60
4.3.1.2 Linearization of power flow and line flow ................................................................ 61
4.3.1.3 Linearization of quadratic function ........................................................................... 63
4.3.2 MILP Model ..................................................................................................................... 65
4.4 Numerical Results ................................................................................................................... 66
4.4.1 Optimum Operation of the Microgrid .............................................................................. 66
4.4.2 Accuracy Evaluation ........................................................................................................ 68
4.4.3 Primary Droop and Secondary Control Impact Evaluation .............................................. 70
4.4.4 Evaluation of Load Shedding Switching State Cycle ...................................................... 72
4.4.5 Evaluation of ESS with primary droop and secondary control ........................................ 73
4.4.5.1 Formulation for ESS with primary droop and secondary control ............................. 73
4.4.5.2 Evaluation of ESS with primary droop and secondary control ................................. 75
4.5 Conclusion............................................................................................................................... 77
Chapter 5 Resilience Oriented Microgrid Planning .................................................................... 78

Résumé .......................................................................................................................................... 78
Abstract ......................................................................................................................................... 78
5.1 Introduction ............................................................................................................................. 79
5.2 Proposed Resilience Oriented Planning Framework ............................................................... 79
5.3 Stage I: Data Preparation......................................................................................................... 82
5.3.1 Distribution System and Planning Elements .................................................................... 82
5.3.2 WT and PV Generation Model......................................................................................... 83
5.3.3 Load Model ...................................................................................................................... 83
5.3.4 Operation Scenarios ......................................................................................................... 84
5.3.4.1 Normal Operation Scenarios ..................................................................................... 84
5.3.4.2 Severe HILP Operation Scenarios ............................................................................. 84
5.4 Stage II: Planning Problem Formulation ................................................................................. 90
viii
5.4.1 Outer Level: Optimal Allocation Level............................................................................ 90
5.4.1.1 Objective Function for Outer Level .............................................................................. 90
5.4.1.2 Constraints for Outer Level ....................................................................................... 93
5.4.1.3 Decision Variables for Outer Level........................................................................... 94
5.4.2 Inner Level: Optimal Power Flow for Microgrid ............................................................. 94
5.4.2.1 Objective Function and Constraints for Inner Level ................................................. 95
5.4.2.2 Decision variables for Inner Level .......................................................................... 101
5.5 Stage III: Optimal Operation based on Deep Neural Network (DNN) learning ................... 101
5.5.1 Set of Sample Inputs and Targets ................................................................................... 102
5.5.2 Microgrid Formation ...................................................................................................... 105
5.5.2.1 Fault Influence Path................................................................................................. 106
5.5.2.2 Fault Influenced Branch and Bus Matrices ............................................................. 107
5.5.2.3 Microgrid Formation Set ......................................................................................... 109
5.5.2.4 Microgrid Formation Indicators .............................................................................. 109
5.5.3 Integration of Microgrid Formation Indicators to Inner Level Optimization ................. 112
5.5.4 Deep Neural Network (DNN) Model ............................................................................. 115
5.6 Stage IV: Solution for Optimum Planning ............................................................................ 117
5.7 Numerical Results ................................................................................................................. 119
5.7.1 Deep Learning Solution Validation................................................................................ 121
5.7.2 NSGA-II Performance based DNN Model .................................................................... 124
5.7.3 Optimal Planning Solution ............................................................................................. 126
5.7.4 Resilience Improvement ................................................................................................. 127
5.8 Conclusion............................................................................................................................. 130
Conclusion and Perspectives ........................................................................................................ 131

6.1 Summary and Conclusions .................................................................................................... 131
6.2 Perspective for the Future Work ........................................................................................... 133
Bibliography .................................................................................................................................. 135
Appendix A: Element of square matrix ...................................................................................... 148
Appendix B: Wind, PV and load models..................................................................................... 150
Appendix C: Algorithm & Example for MGs Formation ......................................................... 151

C.1 Algorithm for MGs Formation ............................................................................................. 151
C.2 Mathematical example for MGs Formation ......................................................................... 154
C.2.1 Fault influence path ....................................................................................................... 154
C.2.2 Branch affected matrix (LFa) ........................................................................................ 156
C.2.4 Bus affected matrix (BFa) ............................................................................................. 157
C.2.5 Microgrid Formation set ................................................................................................ 159
C.2.6 Non-Energized system buses ......................................................................................... 159
Appendix D: Feasible operating regions of DDGs ..................................................................... 161
Appendix E: Linearization of absolute value operator .............................................................. 163

E.1 Linearization of Absolute Value Operator ............................................................................ 163
E.2 Linearize the production between continuous and binary variables ..................................... 163
Appendix F: Decision variables for OPF/LOPF problem ......................................................... 164
ix
Appendix G: Data information .................................................................................................... 165
G.1 Generator’s information ....................................................................................................... 165
G.2 Data information exclusive for Chapter 4 ............................................................................ 166
G.3 Data information exclusive for Chapter 5 ............................................................................ 169
x
List of Figures
Introduction:
FIGURE 0. 1: Basic outline of the research objectives .................................................................... 4
Chapter 1:
FIGURE 1. 1: Different features of typical PDS(a) and ADS(b) ..................................................... 7
FIGURE 1. 2: General scheme of the basic microgrid ..................................................................... 8
FIGURE 1. 3: Steady state, fundamental model of a DG unit operating in the droop mode ........... 9
FIGURE 1. 4: Static Droop characteristics, 𝒇/𝑷 (a) and 𝑽/𝑸 (b) ................................................. 10
FIGURE 1. 5: Droop sharing of two DG units in the microgrid .................................................... 11
FIGURE 1. 6: Bus representation and its variables ........................................................................ 12
Chapter 2:
FIGURE 2. 1: Flowchart of the proposed algorithm ...................................................................... 26
FIGURE 2. 2: Flowchart of the proposed algorithm for unbalanced system ................................. 30
FIGURE 2. 3: Voltage profile for 33-bus IMG system as calculated by NTR [15], MBF [29]
and the proposed algorithm (𝑓= 0.9198 p.u.) .......................................................... 35
FIGURE 2. 4: Comparison between the convergence characteristics of proposed algorithm and
the MBF algorithm of [29] in terms of inner and outer loops error for the 33-bus
IMG system ............................................................................................................. 36
FIGURE 2. 5: Evolution of inner and outer loops error of the proposed algorithm for the 33-
bus IMG system ....................................................................................................... 36
FIGURE 2. 6: Voltage profile of 10476-bus IMG system (𝑓= 0.9952 p.u.) .................................. 37
FIGURE 2. 7: Voltage profile for 25-bus unbalanced IMG as calculated by NTR [15] and by
the proposed algorithm (𝑓= 0.9978 p.u.): (a) Phase A, (b) Phase B and (c) Phase
C .............................................................................................................................. 38
FIGURE 2. 8: Evolution of inner, middle and outer loops error of the proposed algorithm for
the base case 25-bus system .................................................................................... 39
Chapter 3:
FIGURE 3. 1: Voltage profile for 33-bus AMG system as calculated by MBF [29], NTR [15]
and the proposed LPF (𝑓 𝑀𝐵𝐹 ≈ 𝑓 𝑁𝑇𝑅 = 0.9198 and 𝑓 𝐿𝑃𝐹 = 0.9208) ........................ 47
FIGURE 3. 2: Voltage profile for unbalanced 25-bus unbalanced IMG as calculated by NTR
[15] and by the proposed LPF (𝑓 𝑁𝑇𝑅,3Ф = 0.9978 and 𝑓 𝐿𝑃𝐹,3Ф = 0.9981): (a)
Phase A, (b) Phase B and (c) Phase C ..................................................................... 48
FIGURE 3. 3: Voltage profile of 10476-bus AMG system (𝑓= 0.9952)........................................ 48
xi
Chapter 4:
FIGURE 4. 1: Basic scheme of the microgrid ................................................................................ 52
FIGURE 4. 2: (a) Basic scheme of the microgrid (b) Steady-state model of DDG unit’s
operation with primary and secondary control under IS mode................................ 55
FIGURE 4. 3: Feasible operation region of the DDG units in IS and GC mode ............................ 57
FIGURE 4. 4: Principle of the ESS units ....................................................................................... 58
FIGURE 4. 5: Polygon piecewise linearization plan for the quadratic function approximation .... 64
FIGURE 4. 6: Total active power profile for the 25-bus test system: (a) SCEN 1 (b) SCEN 2
(c) SCEN 3 .............................................................................................................. 67
FIGURE 4. 7: Voltage profile at 7:00 (IS mode) and at 8:00 (GC mode) for the unbalanced 33-
bus test case obtained by NPL and MILP................................................................ 69
bus test case obtained by NPL and MILP: (a) Phase A, (b) Phase B, (c) Phase C .. 69
FIGURE 4. 9: Daily voltage magnitude profile of bus #25 obtained by the MILP2nd and
MILPdroop for the 33-bus test system SCEN 2 ......................................................... 71
FIGURE 4. 10: Voltage magnitude profile with and without secondary control for the 33-bus
test system for SCEN 2 at 12:00 am ..................................................................... 71
FIGURE 4. 11: Daily power flowing on branch #24 obtained by MILP2nd and MILPdroop for the
33-bus test system for SCEN 2 ............................................................................. 71
FIGURE 4. 12: Daily system frequency obtained by the MILP2nd and MILPdroop for the 33-bus
test system SCEN 2............................................................................................... 72
FIGURE 4. 13: Number of load shedding switching state cycle .................................................... 73
FIGURE 4. 14: Profile of load shedding at bus #19 ....................................................................... 73
FIGURE 4. 15: Daily profile for (a) binary indicate charging/discharging state of ESS localized
at bus #13 (b) voltage for the bus #13 ................................................................... 76
FIGURE 4. 16: Active discharged power of ESS localized on bus #13 for both case ESS with
and without considering droop+2nd control ......................................................... 76
Chapter 5:
FIGURE 5. 1: Basic outline for the proposed Resilience Oriented Microgrids Planning
framework ............................................................................................................. 81
FIGURE 5. 2: Evolution of network’s system performance during a severe event ....................... 85
FIGURE 5. 3: Generic fragility curve for single pole and conductor against wind gust ............... 86
FIGURE 5. 4: Weather-related algorithm for generating branch availability matrix .................... 89
FIGURE 5. 5: Basic scheme of the single microgrid base ISW sub ................................................ 94
FIGURE 5. 6: Steady state model for DDG unit operate with droop/2nd in IS mode and their
feasible operation region for both GC and IS mode ............................................. 99
FIGURE 5. 7: Proposed optimal operation for hierarchical microgrids based on deep-learning
technique ............................................................................................................. 103
FIGURE 5. 8: Flowchart for generating the microgrid indicator matrices and vectors for time
𝑡 ∈ Ω𝑇 for scenarios s of season se ..................................................................... 105
FIGURE 5. 9: General scheme of multi-layer feed-forward neural network architecture for
predicting a single identified target ..................................................................... 115
FIGURE 5. 10: Optimum planning approach .............................................................................. 118
xii
FIGURE 5. 11: Structure of chromosome for NSGA-II .............................................................. 119
FIGURE 5. 12: Seasonal annual load pattern............................................................................... 120
FIGURE 5. 13: 33-bus test system with candidate buses and branches ....................................... 120
FIGURE 5. 14: MSE over epoch, distributed error of training set and test set of the first DNN
model for predicting 𝑇𝐺1 .................................................................................... 122
FIGURE 5. 17: Comparison multi-objective solutions obtained by proposed NSGA-II based
DNN model and their corresponding real values ................................................ 124
FIGURE 5. 18: Probability density as histogram and cumulative density curve of relative error
𝑂𝐿𝐸𝑉 ∗
𝑒1,𝑅𝐸𝑆𝐼 for 𝑅𝐸𝑆𝐼 𝑂𝐿𝐸𝑉 ......................................................................................... 125
𝑒1,𝑇𝐶 for 𝑇𝐶 𝑂𝐿𝐸𝑉 ............................................................................................. 125
𝑒1,𝐴𝐸𝑀𝐼𝑆 for 𝐴𝐸𝑀𝐼𝑆 𝑂𝐿𝐸𝑉 .................................................................................... 125
FIGURE 5. 21: Classification in Group A and Group B for the multi-objective function’s
solution obtained by NSGA-II based DNN model ............................................. 126
FIGURE 5. 22: Wind gust profile and the timeline of branch #12/#16/#19 availability
corresponding to a selected scenario in spring.................................................... 128
FIGURE 5. 23: Energy supply level and timeline event for representing system performance
for base case, case A1, case A2 and case B2 ...................................................... 129
FIGURE 5. 24: Microgrid formations and bus's supply state at t d for (a) Base case (b) Case A1
(c) Case A2 (d) Case B2 ..................................................................................... 129
Appendix C:
FIGURE C. 1: Path(s) search based on fault at branch #2 ........................................................... 154
FIGURE C. 2: Undirected Graph Tree for 8-bus test system....................................................... 154
FIGURE C. 3: Start, Root and leaves nodes ................................................................................ 155
FIGURE C. 4: Simple 8-bus system with found ISWs and fault at branch #2 ............................ 156
FIGURE C. 5: Buses affected by the original fault at branch #2 ................................................. 158
FIGURE C. 6: 𝐌𝐆𝐬𝐞𝐭 formation................................................................................................... 159
FIGURE C. 7: Microgrids formation and bus no-energize respecting to the location of DDGs . 160
Appendix D:
FIGURE D. 1: Derivation of DDG feasible operation region ...................................................... 161
xiii
List of Tables
Chapter 1:
TABLE 1. 1: Summary of bus classification and variables ........................................................... 12
Chapter 2:
TABLE 2. 1: Proposed algorithm maximum errors compared to MBF [29] and NTR [15] for
the 33-bus balanced and 25-bus unbalanced IMG under different load
characteristics .......................................................................................................... 40
TABLE 2. 2: Average runtime (s) of proposed algorithm, MBF [29] and NTR [15] under
different load characteristics .................................................................................... 40
TABLE 2. 3: Number of iterations of proposed algorithm, MBF [29] and NTR [15] under
different load characteristics .................................................................................... 40
Chapter 3:
TABLE 3. 1: Average runtime (ms) and number of iterations of proposed linear model, MBF
[29] and NTR [15] ................................................................................................... 46
TABLE 3. 2: Proposed LPF error indices compared to MBF [29] and NTR [15] for 33-bus
balanced and 25-bus unbalanced AMG. .................................................................. 47
Chapter 4:
TABLE 4. 1: Summarize of piecewise linear constraint formation for each equipment ............... 65
TABLE 4. 2: Optimization results obtained by MILP for 33-bus and 25-bus in difference
scenarios. ................................................................................................................. 67
TABLE 4. 3: Comparison of results obtained by MILP to those obtained by NPL ....................... 68
TABLE 4. 4: Optimization results for CASE2nd and CASEP ......................................................... 77
Chapter 5:
TABLE 5. 1: Decision Variables Type and Set Belonging for the Outer Level Problem ............. 94
TABLE 5. 2: Decision variables for MILP model for microgrid optimal operation ................... 101
TABLE 5. 3: Hyper-parameters setting for DNN models’ architectures ..................................... 121
TABLE 5. 4: Performance of trained DNN models ..................................................................... 123
TABLE 5. 5: Performance of NSGA-II ....................................................................................... 125
TABLE 5. 6: Planning solution corresponding the max/min in each objective function for
group A and the planning solution for group B ..................................................... 127
TABLE 5. 7: Timeline of overhead line (branch) #12, #16 and #19 break down over the
planning horizon of a selected scenario in spring .................................................. 128
Appendix C:
𝒇𝒂𝒊𝒍
TABLE C. 1: Algorithm for determining 𝑭𝑨𝒃𝒖𝒔 {𝒃𝒓𝒊 }............................................................ 151
TABLE C. 2: Formation of matrix containing microgrid set....................................................... 151
TABLE C. 3: Algorithm for determining 𝒃𝒎𝒕 .............................................................................. 152
TABLE C. 4: Algorithm for forming 𝒃𝒖𝒔𝒏𝒆𝒕 ............................................................................... 153
TABLE C. 5: Algorithm for determining 𝒇𝒃𝒖𝒔 and 𝑫𝑫𝑮𝒅𝒐𝒘𝒏 𝒕 .................................................... 153
xiv
Appendix D:
TABLE D. 2: Constraints and parameters related to the determination of DDG's feasible
operation region ..................................................................................................... 162
Appendix F:
TABLE F. 1: Control and dependent variables of the MINLP and MILP model for minimizing
their objective functions ........................................................................................ 164
Appendix G:
TABLE G. 1: DGs locations and static droop coefficients .......................................................... 165
TABLE G. 2: Daily information data input for the 33-bus and 25-bus test system..................... 166
TABLE G. 3: Load weight and maximum duration of load curtailment 33-bus and 25-bus test
system............................................................................................................................. 166
TABLE G. 4: Characteristic of the DDG, WTs, PVs, ESS, transformer, and transmission line . 167
TABLE G. 5: General parameter information ............................................................................. 168
TABLE G. 6: Set of scenarios and their probability obtained by fast forward selection and
their corresponding calendar day/year for WT and PV .................................................. 169
TABLE G. 7: The most vulnerable wind scenarios and their corresponding date ....................... 169
TABLE G. 8: Parameter for generating scenario ......................................................................... 169
TABLE G. 9: Characteristic of the 33-bus test system ................................................................ 170
TABLE G. 10: General parameters.............................................................................................. 171
xv
Nomenclature
The notation for this thesis has been listed as following. In the order to quick reference, the
parameters, variables, and some of notation type has been stated for each chapter.
Abbreviations
ISW sub Isolate Switch at the upstream substation

AC/DC Alternative/Direct Current system
ADS Active Distribution System
AMG Autonomous or Islanded Microgrid
BFS Backward-Forward-Sweep
BRB Bayesian Regularization Backpropagation
CCS Compressed Column Storage
DDG Dispatchable Distributed Generation
DER Distributed Energy Resource
DFS Depth First Search algorithm
DG Distributed Generation
DMS Demand Side Management
DNN Deep Neural Network
DOE U.S. Department of Energy
droop/2nd Primary droop mode with a supervisory secondary control layer
DSO Distributed System Operator
ESS Energy Storage System unit
GC Grid-Connected operation
GS Gauss Seidel
HILP High Impact Low Probability
IMG Islanded Microgrid
IS Islanded operation
ISW Isolate Switch
LOPF AC,DL Deep Learning AC-LOPF
LPF Linear Power Flow
MBF Modified Backward Forward power flow method
MG(s) Microgrid(s)
MILP Mixed Integer Linear Programming
MINLP Mixed Integer Nonlinear Programming
NaS Technology Sodium Sulfur
N-K Level of contingencies
non-LPF Nonlinear Power Flow
NR Newton Raphson
NREL National Renewable Energy Laboratory
xvi
NSGA-II Non-dominated Sorting Genetic Algorithm II
NTR Newton Trust Region power flow method
NYISO New York Independent System Operator
O&M Operation and Maintenance
OPF/LOPF Non-Linear/Linear Optimum Power Flow
PCC Point of Common Coupling
PHEVs Plug-In Hybrid Electric Vehicles
PQ bus Active and reactive power controllable bus
PQ mode Active and Reactive power control mode
PV Photovoltaic energy
PV bus Active power and voltage controllable bus
SOC State of Charge of the ESS
SPF System Performance
SRF-PLL Synchronous Reference Frame Phase Locked-Loop
T&D Transmission and Distribution
TR Transformer
WT Wind Turbine energy
Sets and Numbers

𝑛𝑏𝑟 Number of the branches in the test system
𝑛𝑏𝑢𝑠 Number of buses in the test system
𝑛𝑑𝑑𝑔 Number of DDGs in the planning problem
𝑛𝐷𝐺 Number of DG units
𝑛𝑒𝑠𝑠 Number of ESSs in the planning problem
𝑛𝑓𝑒𝑎𝑡 Number of features in a vector of sample
𝑛ℎ𝑙 Number of hidden layers for DNN model
𝑛𝑖𝑠𝑤 Number of ISWs in the planning problem
𝑛𝑀𝐺 Number of the microgrids
𝑛𝑝𝑎𝑡ℎ Number of the paths search
𝑛𝑝𝑣 Number of PVs in the planning problem
𝑛𝑝𝑤 Number of pieces for piecewise linearization
𝑁𝑠′ Number of normal scenarios
𝑁𝑠′′ Number of severe scenarios
𝑛𝑠𝑎𝑚 Number of samples for training DNN model
𝑁𝑇 Planning horizon time e.g., 24h
𝑛𝑤𝑡 Number of WTs in the planning problem
ℝ Continuous variable or real number
ℤ Integer variable or binary number
Ω𝑥 Set of buses type x, where x can be {BUS, LIN, PCC, DDG, WDG, PV or ESS}
Ω𝑐𝑎𝑛
𝐷𝐷𝐺 Set of candidate buses for DDGs
Ω𝑐𝑎𝑛
𝐸𝑆𝑆 Set of candidate buses for ESSs
Ω𝑐𝑎𝑛
𝐼𝑆𝑊 Set of candidate branches for SSWs
xvii
Ω𝑐𝑎𝑛
𝑃𝑉 Set of candidate buses for PV units
Ω𝑐𝑎𝑛
𝑊𝑇 Set of candidate buses for WT units
𝑓𝑎𝑖𝑙 Set of failure branches in the test system
Ω𝐿𝐼𝑁
Ω𝑛𝑜𝑑𝑒
𝑇𝐺 Set of targets node(s)
Ω′𝑠 Set of generated scenarios for normal situation
Ω′′
𝑠 Set of generated scenarios for severe situation considering network SPF
progression phase I and II (i.e., 𝑡 𝑒 → 𝑡 𝑟 )
′′,𝑜𝑟𝑖𝑔 Original set before reduction of generated scenarios for severe situation
Ω𝑠
considering network SPF progression phase I and II (i.e., 𝑡 𝑒 → 𝑡 𝑟 )
𝑝𝑟𝑒
Ω𝑠 Set of generated pre-scenarios for severe situation considering only network
SPF progression phase I (i.e., 𝑡 𝑒 → 𝑡 𝑑 )
Ω𝑆𝐴𝑀 Set of DNN training samples
Ω𝑠𝑒 Set of seasonal periods i.e. {fall, winter, spring, summer}
Ω𝑇 Set of segments or hours in the planning horizon e.g. Ω𝑇 = {1,2,3, … ,24}
𝕯 Set of droop buses
ℱ Set of electrical phases i.e. A, B and C
List of Symbols Used in Chapter 1

𝑓 System frequency
𝑓∗ Frequency reference for primary control of the droop unit
𝑓 𝑚𝑎𝑥 /𝑓 𝑚𝑖𝑛 Maximum/Minimum frequency violation
𝑓 𝑂𝑃𝐹 (𝐱) Objective function for OPF problem dependent on set of variables 𝐱
𝐹 𝑝𝑙𝑎𝑛 (𝐱) Objective function of the planning problem dependent on set of variables 𝐱
𝐠 𝐎𝐏𝐅 (𝐱) Set of equality constraints dependent on set of variables 𝐱
𝐆 𝐩𝐥𝐚𝐧 (. ) Set equality and inequality constraint for the planning problem
𝐡𝐎𝐏𝐅 (𝐱) Set of inequality constraints dependent on set of variables 𝐱
𝐇 𝐩𝐥𝐚𝐧 (. ) Set inequality and inequality constraint for the planning problem
̅𝑰 Vector containing current injection in complex form for the buses in the
balanced IMG
𝑚𝑝 Active droop gain of the droop unit
𝑛𝑞 Reactive droop gain of the droop unit
𝑂𝐹𝑚 𝑚𝑡ℎ objective function for the case multi-objective problem
𝑃𝐺,𝑚𝑎𝑥 Maximum allowable for active power generated by DDG unit
𝑃𝐺 Active power generated by the droop unit
𝑃𝑖 Active power injected to bus 𝑖
𝑃𝑖𝐿 Active power demand located at bus 𝑖
𝑄 𝐺,𝑚𝑎𝑥 Maximum allowable for reactive power generated by DDG unit
𝑄𝐺 Reactive power generated by the droop unit
𝑄𝑖 Reactive power injected to bus 𝑖
𝑄𝑖𝐿 Reactive power demand located at bus 𝑖
𝑆𝑖̅ Complex form appearance power injected to bus 𝑖
xviii
𝐮𝐎𝐏𝐅 Set control variables
𝑽̅ Vector containing voltage in complex form for the buses in the balanced IMG
|𝑉̅ | Magnitude of voltage
𝑉∗ Voltage reference for primary control of droop unit
𝑉 𝑚𝑎𝑥 /𝑉 𝑚𝑖𝑛 Maximum/Minimum voltage violation
𝐱 𝐎𝐏𝐅 Set of dependent variables
𝐱 𝐩𝐥𝐚𝐧 Set decision variables for the planning problem
𝐘̅ 𝐛𝐮𝐬 Admittance matrix in rectangular form for the balanced IMG
𝐙̅𝐛𝐮𝐬 Impedance matrix in rectangular form for the balanced IMG
𝛿 Phase angle of voltage
𝜔 Angular frequency
Parameters:
𝐿 Active power demand phase 𝛹 located at bus 𝑖
𝑃𝑖,𝛹
𝑃𝑖𝐿,3Ф Three phase active power demand located at bus 𝑖
𝑃𝑖𝐿 Active power demand located at bus 𝑖
𝐿 Reactive power demand phase 𝛹 located at bus 𝑖
𝑄𝑖,𝛹
𝑄𝑖𝐿,3Ф Three phase reactive power demand located at bus 𝑖
𝑄𝑖𝐿 Reactive power demand located at bus 𝑖
𝑉𝑖∗ Voltage reference for primary control of the 𝑖 𝑡ℎ DDG
𝑓𝑖∗ Frequency reference for primary control of the 𝑖 𝑡ℎ DDG
𝑝
𝑚𝑖 Active droop gain of the 𝑖 𝑡ℎ DDG unit
𝑞
𝑛𝑖 Reactive droop gain of the 𝑖 𝑡ℎ DDG unit
Variables and elements exclusive used for chapter 2:

∆𝑓 Frequency deviation of the IMG from the nominal frequency
̅̅̅̅
𝒅𝑽 Vector containing voltage deviation for the buses in the balanced IMG
̅̅̅̅
𝑑𝑉𝑖 Voltage deviation calculated for bus 𝑖
𝑓 Islanded Microgrid’s frequency
𝑓0 System frequency of nominal operating point
𝑓 3Ф Islanded Microgrid’s frequency considering 3 phase power flow
𝑰̅ Vector containing current injection in complex form for the buses in the
balanced IMG
𝑰̅𝒃𝒓,𝒂𝒃𝒄 Vector containing complex form of branch current phase A, B and C for the
unbalanced IMG
̅𝑰𝒂𝒃𝒄 Vector containing complex form of current injected to phase A, B and C for the
buses in the unbalanced IMG
̅𝑰𝒃𝒓 Vector containing complex form of branch current in the unbalanced IMG
̅
𝑏𝑟
𝐼1,𝛹 Sum of branch current phase 𝛹 that adjacent to bus 1
̅
𝐼𝑖,𝛹 Current injected to phase 𝛹 of bus 𝑖
xix
𝑏𝑟
̅ Current flowing on phase 𝛹 of branch 𝑖
𝐼𝑖,𝛹
𝐼1̅
𝑏𝑟 Sum of branch current that adjacent to bus 1
𝐼𝑖̅ Current injected to bus 𝑖
𝐼𝑖𝑏𝑟 ̅ Current of branch 𝑖
𝑖𝑡𝑀 Iteration counter of the middle loop of Zbus algorithm
𝐊/𝐊 𝐓 Incidence matrix and it transpose of the unbalanced IMG
𝐊 𝟑Ф /(𝐊 𝟑Ф )T Three phase incidence matrix and it transpose of the unbalanced IMG
𝐾 𝑝𝑓 /𝐾 𝑞𝑓 Active and reactive power frequency dependencies factors
𝑃𝑖,𝛹 Active power injected phase 𝛹 to bus 𝑖
𝐺,𝑖𝑛𝑖𝑡
𝑃𝑖,𝛹 Initial active power generated phase 𝛹 by 𝑖 𝑡ℎ DG unit
𝐿,𝑠𝑝𝑒𝑐 Specified nominal active power demand
𝑃𝑗,𝛹
𝑃𝑖𝑎,𝑏,𝑐 Active power injected specific to phase A, B or C of bus 𝑖
𝑃𝑖𝑏𝑟,3Ф Three-phase active power flowing on branch 𝑖
𝑃𝑖
𝐺,(𝑎,𝑏,𝑐) Active power generated phase A, B or C by the 𝑖 𝑡ℎ DG unit
𝑃𝑖𝐺,3Ф Three phase active power generated by the 𝑖 𝑡ℎ DG unit
𝑃𝑖𝐺 Active power generated by the 𝑖 𝑡ℎ DG unit
𝑄 𝐺𝑖 Reactive power generated by the 𝑖 𝑡ℎ DG unit
𝐺,𝑖𝑛𝑖𝑡
𝑄𝑖,𝛹 Initial reactive power generated phase 𝛹 by 𝑖 𝑡ℎ DG unit
𝐿,𝑠𝑝𝑒𝑐 Specified nominal reactive power demand
𝑄𝑗,𝛹
𝑄𝑖𝑎,𝑏,𝑐 Reactive power injected specific to phase A, B or C of bus 𝑖
𝑄𝑖𝑏𝑟,3Ф Three-phase reactive power flowing on branch 𝑖
𝑄𝑖
𝐺,(𝑎,𝑏,𝑐) Reactive power generated phase A, B or C by the 𝑖 𝑡ℎ DG unit
𝑄𝑖𝐺,3Ф Three phase reactive power generated by the 𝑖 𝑡ℎ DG unit
𝑄𝑖𝐿,3Ф Three phase reactive power demand located at bus 𝑖
𝐑 𝐛𝐮𝐬
Real part of 𝐙̅𝐛𝐮𝐬
̅
𝑺 Vector containing power injection in complex form for the buses in the balanced
IMG
̅𝒂𝒃𝒄
𝑺 Vector containing complex form of power injected to phase A, B and C for the
buses in the unbalanced IMG
𝑽 Vector containing voltage magnitude in for the buses in the balanced IMG
̅
𝑽 Vector containing voltage in complex form for the buses in the balanced IMG
̅ 𝒂𝒃𝒄
𝑽 Vector containing the voltages in complex form for the buses in the unbalanced
IMG
𝑉̅1,𝛹 Voltage in complex form for phase 𝛹 of bus 1
𝑉̅𝑖,𝛹 Voltage in complex form for phase 𝛹 of bus 𝑖
̅ 𝒂𝒃𝒄
𝑽 𝟏 Vector containing voltage in complex form for phase A, B and C of bus 1
𝑉̅𝑖 Voltage in complex form of bus 𝑖
𝑉̅𝑖𝑎,𝑏,𝑐 Voltage in complex form specific to phase A, B and C of bus 𝑖
xx
𝑽𝑖𝑡 Vector of iteration 𝑖𝑡, containing voltage magnitude for the buses in the balanced
IMG
𝑉1 Voltage magnitude of bus 1
𝑉1𝑖𝑡 Voltage magnitude of bus 1 for iteration 𝑖𝑡
𝑉𝑖 Voltage magnitude of the bus 𝑖
𝐗 𝐛𝐫,𝟑Ф Reactance matrix for the branches in the unbalanced IMG
𝐗 𝐛𝐮𝐬,(𝟎) Reactance matrix for the buses in the balanced IMG based on initial data import
𝐗 𝐛𝐮𝐬,𝟑Ф Reactance matrix for the buses in the unbalanced IMG
𝐗 𝐛𝐮𝐬 Reactance matrix for the buses in the balanced IMG
𝛹𝛽 Admittance between node 𝑖 and 𝑗 considering phase 𝛹 to 𝛽
𝑌𝑖𝑗
𝐙̅𝐛𝐫,𝟑Ф Primitive impedance matrix in rectangular form of the unbalanced IMG
𝐙̅𝐛𝐮𝐬,𝟑Ф Three-phase bus impedance matrix in rectangular form for unbalanced IMG
𝐙̅𝐛𝐮𝐬 Impedance matrix in rectangular form for the balanced IMG
̅ 𝑏𝑟,𝛹𝛽
𝑍𝑖𝑗 Impedance between node 𝑖 and 𝑗 considering phase 𝛹 to 𝛽
𝛼/𝛽 Active and reactive power exponents
𝜹 Vector containing voltage magnitude angle in for the buses in the balanced IMG
𝜹𝒂𝒃𝒄 Vector containing the voltage magnitude angles for the buses in the unbalanced
IMG
𝛿𝑖 Voltage magnitude angle of the bus 𝑖
𝛿𝑖𝑎,𝑏,𝑐 Voltage magnitude angle specific to phase A, B or C of bus 𝑖
𝛹𝛽 Admittance angle between node 𝑖 and 𝑗 considering phase 𝛹 to 𝛽
𝜃𝑖𝑗
𝜆 Step size exclusive used for the middle of Zbus algorithm
Parameters:
𝐿 Active power demand phase 𝛹 located at bus 𝑖
𝑃𝑖,𝛹
𝐿 Reactive power demand phase 𝛹 located at bus 𝑖
𝑄𝑖,𝛹
𝑉𝑖∗ Voltage reference for primary control of the 𝑖 𝑡ℎ DDG
𝑓𝑖∗ Frequency reference for primary control of the 𝑖 𝑡ℎ DDG
𝑝
𝑚𝑖 Active droop gain of the 𝑖 𝑡ℎ DDG unit
𝑞
𝑛𝑖 Reactive droop gain of the 𝑖 𝑡ℎ DDG unit
Variables:
𝛹𝛽 Element of the susceptance matrix denote the connection of bus 𝑖 phase 𝛹 to
𝐵𝑖𝑗
bus 𝑗 phase 𝛽
𝑓 3Ф Islanded Microgrid’s frequency considering 3 phase power flow
𝛹𝛽 Element of the conductance matrix denote the connection of bus 𝑖 phase 𝛹 to
𝐺𝑖𝑗
bus 𝑗 phase 𝛽
𝑃𝑖,𝛹 Active power injected phase 𝛹 to bus 𝑖
𝐺
𝑃𝑖,𝛹 Active power generated by the 𝑖 𝑡ℎ DG unit
xxi
𝑃𝑖𝐺,3Ф Three phase active power generated by the 𝑖 𝑡ℎ DG unit
𝑄𝑖,𝛹 Reactive power injected phase 𝛹 to bus 𝑖
𝑄𝑖𝐺,3Ф Three phase reactive power generated by the 𝑖 𝑡ℎ DG unit
𝑉̅𝑖,𝛹 Voltage magnitude in complex form of phase 𝛹 of the bus 𝑖
𝑉̅𝑖𝑎,𝑏,𝑐 Voltage magnitude in complex form specific to phase A, B or C of the bus 𝑖
𝛿𝑖,𝛹 Voltage magnitude angle of phase 𝛹 of the bus 𝑖
𝛿𝑖𝑎,𝑏,𝑐 Voltage magnitude angle of specific to phase A, B or C of the bus 𝑖
Parameters and vector/matrix:

∆𝑡 Step time (i.e., 1 segment = 1 hour)
𝐴𝑃𝐿𝑡 Annual peak load factor according to seasonal pattern
𝛹𝛽 Element of the susceptance matrix denote the connection of bus 𝑖 phase 𝛹 to
𝐵𝑖𝑗
bus 𝑗 phase 𝛽
𝑏𝑡𝑚 Binary input parameter indicates the operation mode of the MG at time 𝑡
𝑏𝐸𝑆𝑆𝑚
𝑖 Binary input parameter indicates the operation mode of the 𝑖 𝑡ℎ ESS
𝐶𝑅𝑖𝐷𝐷𝐺 Consumption rate of the natural gas of DDGi [scf/kWh]
𝐸𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 Maximum allowable stored energy of the ESS unit 𝑖
𝐸𝑖𝐸𝑆𝑆,𝑚𝑖𝑛 Minimum allowable stored energy of the ESS unit 𝑖
𝛹𝛽 Element of the conductance matrix denote the connection of bus 𝑖 phase 𝛹 to
𝐺𝑖𝑗
bus 𝑗 phase 𝛽
𝐼𝑘𝑏𝑟,𝑚𝑎𝑥 Maximum ampacity of the branch 𝑘
𝑘𝑖 𝜔 Integral gain for frequency restoration for DDG or for ESS with primary droop
control mode with a supervisory secondary control layer
𝑝
𝑚𝑖 Active droop gain of the 𝑖 𝑡ℎ DDG unit or the 𝑖 𝑡ℎ ESS with primary droop control
mode with a supervisory secondary control layer
𝑞
𝑛𝑖 Reactive droop gain of the 𝑖 𝑡ℎ DDG unit or the 𝑖 𝑡ℎ ESS with primary droop
control mode with a supervisory secondary control layer
𝑁𝑐𝑢𝑟𝑡
𝑖 Limit on the number of load shedding switchings for each load at bus 𝑖
𝐿,3𝛷 Three phase active power demand located at bus 𝑖, time 𝑡
𝑃𝑖,𝑡
𝑃𝑉,3𝛷
𝑃𝑖,𝑡 Three phase active power generated by 𝑖 𝑡ℎ PV units at time 𝑖
𝑊𝑇,3𝛷
𝑃𝑖,𝑡 Three phase active power generated by 𝑖 𝑡ℎ wind units at time 𝑖
𝐿,𝑝𝑒𝑎𝑘 Annual peak active demand of bus 𝑖
𝑃𝑖,𝛹
𝐿 Active power demand phase 𝛹 located at bus 𝑖, time 𝑡
𝑃𝑖,𝛹,𝑡
𝑃𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 Maximum allowable for active power generated by 𝑖 𝑡ℎ DDG unit
𝑃𝑖𝐷𝐷𝐺,𝑚𝑖𝑛 Minimum allowable for active power generated by 𝑖 𝑡ℎ DDG unit
𝑉𝐼
𝑃𝐶 Penalty coefficient of voltage violation [$/kV]
𝐷𝐷𝐺,1𝑠𝑡/4𝑡ℎ Minimum operating power factor allowable of the 1st/4th quadrant of the DDG
𝑃𝐹𝑖
unit 𝑖
xxii
𝑇𝑅,1𝑠𝑡/4𝑡ℎ Minimum operating power factor allowable of the 1st/4th quadrant of the
𝑃𝐹𝑖
transformer 𝑖
𝑃𝐹𝑖𝑃𝑉 Power factor set of the PV unit 𝑖
𝑃𝐹𝑖𝑊𝑇 Power factor set of the wind unit
𝑊𝑇,3𝛷
𝑄𝑖,𝑡 Three phase reactive power generated by 𝑖 𝑡ℎ wind units at time 𝑖
𝐿,𝑝𝑒𝑎𝑘 Annual peak reactive demand of bus 𝑖
𝑄𝑖,𝛹
𝐿 Reactive power demand phase 𝛹 located at bus 𝑖, time 𝑡
𝑄𝑖,𝛹,𝑡
𝑄𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 Maximum allowable for reactive power generated by 𝑖 𝑡ℎ DDG unit
𝑄𝑖𝐷𝐷𝐺,𝑚𝑖𝑛 Minimum allowable for reactive power generated by 𝑖 𝑡ℎ DDG unit
𝑄𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 Maximum allowable for reactive power of ith ESS unit
𝑄𝑖𝐸𝑆𝑆,𝑚𝑖𝑛 Minimum allowable for reactive power of ith ESS unit
𝐶 Certain radiation point, typically set to 0.15 kW/m2
𝑅
𝑅 𝑆𝑇𝐷 Solar radiation in standard test conditions, typically specified as 1000 W/m2
𝑟𝑡𝑚 Solar irradiance in kW/m2
𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 Capacity rating of the DDG unit 𝑖
𝑆𝑖𝑃𝑉,𝑚𝑎𝑥 Capacity rate of the PV unit located at bus 𝑖
𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 Capacity rate of the transformer located at the substation
𝑆𝑖𝑊𝑇,𝑚𝑎𝑥 Capacity rate of the wind DG located at bus 𝑖
𝑆𝑂𝐶𝑖0 Initial state of charge of the ESS unit 𝑖
𝑇𝑐𝑢𝑟𝑡
𝑖 Maximum hours of load curtailment of bus 𝑖
𝑐𝑖 𝑐𝑜 𝑟𝑎
𝑣 ,𝑣 ,𝑣 Cut-in, cut-out, and the rated speed of the wind DG, respectively
𝑉 𝑚𝑎𝑥 Maximum voltage violation
𝑉 𝑚𝑖𝑛 Minimum voltage violation
𝑉𝑖∗∗ Voltage reference for secondary control of the 𝑖 𝑡ℎ DDG or the 𝑖 𝑡ℎ ESS with
primary droop control mode with a supervisory secondary control layer
𝑉𝑖∗ Voltage reference for primary control of the 𝑖 𝑡ℎ DDG or the 𝑖 𝑡ℎ ESS with
primary droop control mode with a supervisory secondary control layer
𝐺𝐶
𝑉𝑠𝑝𝑒𝑐 Nominal specified voltage during grid-connected mode
𝑣𝑡𝑚 Wind speed in m/s
𝜂𝑖𝐸𝑆𝑆 Efficiency of the energy storage
𝜎 𝑐𝑢𝑟𝑡 Cost of load curtailment [$/kWh]
𝜎 𝑓𝑢𝑒𝑙 Natural gas price [$/Thousands of cubic feet]
𝜎𝑡𝑃𝐶𝐶 Market energy price at time 𝑡 [$/kWh]
𝜔∗ Nominal angular frequency of the system
Sets of Variables:
𝑿𝑀𝐼𝑁𝐿𝑃,3𝛷 Vector containing the decision variables for the MINLP problem formulation
𝑿𝑀𝐼𝐿𝑃,3𝛷 Vector containing the decision variables for the MILP problem formulation
Continuous variables:
𝐸𝑆𝑆 Energy SOC of the ESS unit 𝑖 at time 𝑡
𝐸𝑖,𝑡
xxiii
𝐷𝐷𝐺,3𝛷
𝑃𝑖,𝑡 Three phase active power generated by the 𝑖 𝑡ℎ DDG unit at time 𝑡
𝑃𝐶𝐶,3𝛷
𝑃𝑖,𝑡 Three phase active power imported from the main grid across the 𝑖 𝑡ℎ PCC of the
MG at time 𝑡
𝑐ℎ,3𝛷
𝑃𝑖,𝑡 Active power charge by 𝑖 𝑡ℎ ESS unit at time 𝑡
𝑑𝑐ℎ,3𝛷
𝑃𝑖,𝑡 Active power discharge by 𝑖 𝑡ℎ ESS unit at time 𝑡
𝑃𝑖,𝛹,𝑡 Active power injected phase 𝛹 to bus 𝑖 at time 𝑡
𝐷𝐷𝐺
𝑃𝑖,𝛹,𝑡 Active power phase 𝛹 generated by the 𝑖 𝑡ℎ DDG unit at time 𝑡
𝑃𝐶𝐶
𝑃𝑖,𝛹,𝑡 Active power phase 𝛹 imported from the main grid across the 𝑖 𝑡ℎ PCC of the
MG at time 𝑡
𝑑𝑐ℎ
𝑃𝑖,𝛹,𝑡 Active power phase 𝛹 discharged by 𝑖 𝑡ℎ ESS unit at time 𝑡
𝑏𝑟 Active power phase 𝛹 flowing across the transmission branch 𝑘 at time 𝑡
𝑃𝑘,𝛹,𝑡
𝐸𝑆𝑆,3𝛷
𝑄𝑖,𝑡 Three reactive power generate/receive of the 𝑖 𝑡ℎ ESS unit, at time 𝑡
𝑄𝑖,𝛹,𝑡 Re-active power injected phase 𝛹 to bus 𝑖 at time 𝑡
𝐷𝐷𝐺
𝑄𝑖,𝛹,𝑡 Reactive power phase 𝛹 generated by the 𝑖 𝑡ℎ DDG unit at time 𝑡
𝐸𝑆𝑆
𝑄𝑖,𝛹,𝑡 Reactive power phase 𝛹 generated by 𝑖 𝑡ℎ ESS unit at time 𝑡
𝑃𝐶𝐶
𝑄𝑖,𝛹,𝑡 Three phase reactive power imported from the main grid across the 𝑖 𝑡ℎ PCC of
the MG at time 𝑡
𝑏𝑟 Reactive power phase 𝛹 flowing across the transmission branch 𝑘 at time 𝑡
𝑄𝑘,𝛹,𝑡
𝑣,𝑖𝑛𝑡
𝑢𝑖,𝛹,𝑡 Integrator output of secondary voltage restoration considering phase 𝛹 of 𝑖 𝑡ℎ
DDG unit or the 𝑖 𝑡ℎ ESS operating with secondary control at time 𝑡
𝑉𝑖,𝛹,𝑡 Voltage magnitude phase 𝛹 , of the bus i at time t
𝛿𝑖,𝛹,𝑡 Voltage magnitude angle of phase 𝛹, bus i at time 𝑡
𝑀𝐺
𝛿𝛹,𝑡 Phase angle of the bus where the frequency control act
Binary variables:
𝑏𝑐𝑢𝑟𝑡
𝑖,𝑡 Binary variable representing load curtailment (1: load shaded and 0: load fed)
𝑏𝐸𝑆𝑆
𝑖,𝑡 Binary variable indicates the status of ESS i.e. 1: discharge and 0: charge
z𝑐𝑢𝑟𝑡
𝑖,𝑡 Binary variable representing a linearization of absolute term |𝑏𝑐𝑢𝑟𝑡 𝑐𝑢𝑟𝑡
𝑖,𝑡 − 𝑏𝑖,𝑡−1 |
Function and Metric:

𝐶𝑡𝐶𝑈𝑅𝑇 Cost of energy curtailment during islanding mode at time 𝑡
𝐶𝑡𝐷𝐷𝐺 Cost of natural gas consumption by the DDG units at time 𝑡
𝐶𝑡𝑃𝐶𝐶 Cost of energy imported from the substation across the PCC of the microgrid at
time 𝑡
𝐶𝑡𝑉𝐼 Cost of voltage deviation at time 𝑡
𝐶𝑡𝑉𝐼,𝐿𝑖𝑛 Linearized version of 𝐶𝑡𝑉𝐼
Parameters and vector/matrix:
xxiv
∆𝑡 Step time (i.e., 1 segment = 1 hour)
𝐀𝐌𝐆 Adjacency matrix for generating 𝐌𝐆𝑡𝐬𝐞𝐭
𝐴𝑃𝐿𝑡,𝑠,𝑠𝑒 Annual peak load pattern in percentage at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝐀𝐕 Line or branch availability binary matrix
𝐈−𝐈𝐈 𝐈−𝐈𝐈,𝐨𝐫𝐢𝐠
𝐀𝐕𝑠,𝑠𝑒 Reduction version of 𝐀𝐕𝑠,𝑠𝑒
𝐈−𝐈𝐈,𝐨𝐫𝐢𝐠 Matrix constraining the binary element represents an availability of the branches
𝐀𝐕𝑠,𝑠𝑒
for 𝑠 𝑡ℎ scenario of season 𝑠𝑒, considering full network SPF progression i.e.,
phase I and II (i.e., 𝑡 𝑒 → 𝑡 𝑟 )
I
𝐀𝐕𝑠,𝑠𝑒 Matrix constraining the binary element represents an availability of the branches
for 𝑠 𝑡ℎ scenario of season 𝑠𝑒, considering only network SPF progression phase
I (i.e., 𝑡 𝑒 → 𝑡 𝑑 )
𝐵𝑖𝑗,𝑡,𝑠,𝑠𝑒 Element of the time-coupled susceptance matrix denote the connection of bus i
to bus j at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝐁𝑠,𝑠𝑒 3D matrix containing the time coupled susceptance matrices for 𝑠 𝑡ℎ scenario of
season 𝑠𝑒
𝑚
𝑏𝑡,𝑠,𝑠𝑒 Binary input parameter indicates the operation mode of the microgrid at time 𝑡
for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝐛𝑖 Vector of biases for 𝑖 𝑡ℎ hidden layer of DNN architecture
𝐵𝑖𝑗 Element of the susceptance matrix denote the connection of bus 𝑖 to bus 𝑗
𝑚
𝒃𝒎
𝑡 Vector containing binary elements 𝑏𝑖,𝑡,𝑠,𝑠𝑒 indicates the operation mode for each
bus at time 𝑡
𝑩𝑭𝒂𝑡 Binary vector represents the bus(es) affected/non-affected by fault(s) at time 𝑡
𝑓𝑎𝑖𝑙 Branch location corresponding to failure branch 𝑖
𝑏𝑟𝑖
𝒃𝒖𝒔𝒐𝒏 Bus online or bus non affected by fault(s)
𝑛𝑒
𝒃𝒖𝒔𝑛𝑒
𝑡 Vector containing binary elements 𝑏𝑢𝑠𝑖,𝑡,𝑠,𝑠𝑒 that indicates bus energize/non-
energize at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑛𝑜𝑟
𝑐𝑑𝑓 (.) Normal Cumulative Distribution Function
𝑠𝑖𝑡𝑒
𝐷𝐷𝐺𝑛,𝑠𝑝 Location of 𝑛𝑡ℎ DDG of sample 𝑠𝑝
𝑠𝑖𝑡𝑒 Vector containing the location of the DDGs for sample 𝑠𝑝
𝑫𝑫𝑮𝑠𝑝
𝑠𝑖𝑧𝑒 Vector containing the size of DDGs for sample 𝑠𝑝
𝑫𝑫𝑮𝑠𝑝
𝑫𝑫𝑮𝑑𝑜𝑤𝑛
𝑡 Vector containing binary elements 𝐷𝐷𝐺𝑛,𝑡 𝑑𝑜𝑤𝑛
that indicate the 𝑛𝑡ℎ DDG
deactivation at time 𝑡
𝐷𝐼𝐽𝐾(. ) Shortest path between two points by Dijkstra algorithm
𝐷𝑜𝐷 𝑚𝑎𝑥 Maximum Depth of Discharge allowable for ESS
𝐸𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 Energy rating of ESS at candidate bus 𝑖
𝑠𝑖𝑡𝑒 Vector containing the location of the ESSs for sample 𝑠𝑝
𝑬𝑺𝑺𝑠𝑝
𝑬𝑺𝑺𝑑𝑜𝑤𝑛
𝑡
𝑑𝑜𝑤𝑛
Vector containing the status of the ESSs (i.e., 𝐸𝑆𝑆𝑖,𝑡,𝑠,𝑠𝑒 ) at time 𝑡
′
𝑓 𝑑𝑛𝑛 Activation function for the output layer
𝑓𝑖𝑑𝑛𝑛 (.) Activation function for the 𝑖 𝑡ℎ hidden layer
𝑏𝑢𝑠
𝒇𝒃𝒖𝒔
𝑡 Vector containing 𝑓𝑛,𝑡 that indicate the location of the controllable bus at time
𝑡ℎ
𝑡 for a 𝑛 DDG
xxv
𝐅𝐀𝐛𝐫 Binary matrix represents branches affected by different location of faults
𝐅𝐀𝐛𝐮𝐬 Binary matrix represents buses affected by different location of faults
𝑓𝑎𝑖𝑙 From bus of the failure overhead line of branch 𝑖
𝑓𝑏(𝑏𝑟𝑖 )
𝒇𝒃, 𝒕𝒃 Set of from bus and to bus for the test case tropology
𝒇𝒍𝒐𝒄𝑡 Vector containing the 𝑖 𝑡ℎ fault at location i.e., 𝑓𝑙𝑜𝑐𝑖,𝑡 at time 𝑡
𝐆𝐛𝐮𝐬 Undirected graph representation by set of vertices (nodes) 𝑵𝑫𝒃𝒖𝒔 and set of
edge (branch) 𝑬𝑫𝒃𝒓
𝐺𝑖𝑗,𝑡,𝑠,𝑠𝑒 Element of the time-coupled conductance matrix denote the connection of bus 𝑖
to bus 𝑗 at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝐆𝑠,𝑠𝑒 3D matrix containing the time coupled conductance matrices for 𝑠 𝑡ℎ scenario of
season 𝑠𝑒
𝐺𝑖𝑗 Element of the conductance matrix denote the connection of bus 𝑖 to bus 𝑗
𝑑𝑑𝑔
𝒉𝒅𝒅𝒈 Vector containing ℎ𝑚 that indicate the presents of DDG for 𝑚𝑡ℎ MG
ℎ𝑖𝑔ℎ
ℎ1 /ℎ2
ℎ𝑖𝑔ℎ Pre-determined range for determining 𝜑𝑏𝑟 for the case corresponding to higher
𝑔𝑢𝑠𝑡 ℎ𝑖𝑔ℎ
wind 𝑣𝑡,𝑠,𝑠𝑒 > 𝑣𝑐𝑟𝑖
ℎ1𝑙𝑜𝑤 /ℎ2𝑙𝑜𝑤 𝑙𝑜𝑤
Pre-determined range for determining 𝜑𝑏𝑟 for the case corresponding to 𝑣𝑐𝑟𝑖 ≤
𝑔𝑢𝑠𝑡 ℎ𝑖𝑔ℎ
𝑣𝑡,𝑠,𝑠𝑒 ≤ 𝑣𝑐𝑟𝑖
𝐼𝑘𝑏𝑟,𝑚𝑎𝑥 Maximum ampacity of the branch 𝑘
𝐈𝐍 𝐝𝐧𝐧 Matrix containing vector of feature 𝑰𝑵𝒅𝒏𝒏
𝑠𝑝 for training DNN model, 𝑠𝑝 ∈ Ω𝑆𝐴𝑀
𝑜𝑝𝑒𝑛
𝑰𝑺𝑾𝑜𝑝𝑒𝑛 Vector containing the location of mth ISW (i.e., 𝐼𝑆𝑊𝑚 ) in open state
𝑰𝑺𝑾𝑠𝑖𝑡𝑒 Vector containing the location of ISWs
𝑘 𝑔𝑢𝑠𝑡 Constant factor for calculating wind-gust
𝑘 𝑖𝜔 Integral gain for frequency restoration for DDG with droop/2nd
𝐻𝐼
𝐾𝑠,𝑠𝑒 /𝐾𝑠,𝑠𝑒 Level of contingencies
𝐷𝐷𝐺
𝐾𝐶𝑂2 Carbon emission factor in kg/kWh according to energy generated by DDGs
𝐺𝑅𝐼𝐷
𝐾𝐶𝑂2 Carbon emission factor kg/kWh according to energy imported from the main
grid
𝐿𝐷𝐷𝐺
𝑖,𝑡,𝑠,𝑠𝑒 Set of piecewise linearization for DDG quadratic constraint
𝐿𝐸𝑆
𝑖,𝑡,𝑠,𝑠𝑒 Set of piecewise linearization for ESS quadratic constraint
𝐿𝑃𝐶𝐶
𝑖,𝑡,𝑠,𝑠𝑒 Set of piecewise linearization for PCC quadratic constraint
𝑏𝑟 Set of piecewise linearization for line ampacity quadratic constraint
𝐿𝑘,𝑡,𝑠,𝑠𝑒
𝑳𝑭𝒂𝑡 Binary vector represents the line(s) affected/non-affected by fault(s) at time t
𝐌𝐚𝐝𝐣 Adjacency matrix
𝑝
𝑚𝑖 Active droop gain of the 𝑖 𝑡ℎ DDG unit with primary droop/2nd
𝑎𝑑𝑗
𝑀𝑖𝑗 Binary element of 𝐌 𝐚𝐝𝐣 represents the connection between two adjacent nodes
𝑖𝑗
𝑴𝑮𝑫𝑭𝑺
𝑚 Vector of 𝑚𝑡ℎ microgrid obtained by DFS of 𝑚𝑡ℎ 𝑟𝑜𝑜𝑡 𝑀𝐺
𝐌𝐆𝐬𝐞𝐭𝟎 Initial set of microgrid formation
𝑴𝑮𝒔𝒆𝒕
𝑚,𝑡 Set of buses in 𝑚𝑡ℎ MG formed at time t
𝐌𝐆𝑡𝐬𝐞𝐭 Matrix containing the set of possible microgrids formation due to fault at time t
xxvi
MSE Mean sum of squares of the DNN errors
MSW Mean sum of squares of the DNN weights
𝑀𝑇𝑇𝑅 𝑛𝑤 Specified mean time to repair under normal weather condition
𝑝𝑜𝑙𝑒 Number of the poles according to span length of branch 𝑖
𝑁𝑖
𝑞
𝑛𝑖 Reactive droop gain of the 𝑖 𝑡ℎ DDG unit with primary droop/2nd
𝑝𝑟𝑒 𝑝𝑟𝑒
𝑁𝑠 Number of generated scenarios in Ω𝑠
𝑛 (. )
𝑛𝑒𝑡 Trained DNN model for target 𝑛
𝑃𝐷𝐷𝐺,𝑚𝑎𝑥 Planning upper bound for DDGs allocation
𝑃𝐸𝑆𝑆,𝑚𝑎𝑥 Planning upper bound for ESSs allocation
𝑃𝑃𝑉,𝑚𝑎𝑥 Planning upper bound for PVs allocation
𝑃𝑊𝑇,𝑚𝑎𝑥 Planning upper bound for WTs allocation
𝒑𝒅𝒏𝒏
𝑖,𝑠𝑝 Vector of neuron pre-activation obtained by 𝑖 𝑡ℎ hidden layer according to
sample 𝑠𝑝
𝑃𝑖,𝑡,𝑠,𝑠𝑒 Active power injected to bus 𝑖 at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝐿
𝑃𝑖,𝑡,𝑠,𝑠𝑒 Active power load for bus 𝑖 at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑃𝑉
𝑃𝑖,𝑡,𝑠,𝑠𝑒 Active power generated by 𝑖 𝑡ℎ PV unit at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑊𝑇
𝑃𝑖,𝑡,𝑠,𝑠𝑒 Active power generated by 𝑖 𝑡ℎ WT unit at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠
𝐺𝑅𝐼𝐷+
𝑃𝑡,𝑠,𝑠𝑒 Active power imported from the main grid at time 𝑡 for 𝑠 𝑡ℎ scenario of season
𝑃𝐶𝐶
𝑠𝑒. Its corresponding to 𝑃𝑖=1,𝑡,𝑠,𝑠𝑒 >0
𝑃𝑖𝐷𝐷𝐺,𝑠𝑖𝑧𝑒 Active power rating of DDG at candidate bus 𝑖
𝑃𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 Active power rating of ESS at candidate bus 𝑖
𝐿,𝑝𝑒𝑎𝑘
𝑃𝑖,𝑡,𝑠,𝑠𝑒 Original active peal load from data sheet, bus 𝑖 at time t for 𝑠 𝑡ℎ scenario of
season 𝑠𝑒
𝑃𝑖𝑃𝑉,𝑠𝑖𝑧𝑒 Active power rating of PV at candidate bus 𝑖
𝑃𝑖𝑊𝑇,𝑠𝑖𝑧𝑒 Active power rating of WT at candidate bus 𝑖
𝑃𝑖𝑏𝑢𝑠,𝑚𝑎𝑥 Maximum installation power allowable for DERs at candidate bus 𝑖
𝐷𝐷𝐺
𝑃𝑠𝑡𝑒𝑝 Step size for DDG allocation
𝐸𝑆𝑆
𝑃𝑠𝑡𝑒𝑝 Step size for ESS allocation
𝑃𝑉
𝑃𝑠𝑡𝑒𝑝 Step size for PV allocation
𝑊𝑇
𝑃𝑠𝑡𝑒𝑝 Step size for WT allocation
𝐏𝐀𝐓𝐇 𝐛𝐫 Matrix containing the vector 𝒑𝒂𝒕𝒉𝒃𝒓 𝑛𝑜𝑑𝑒
𝑖 with ∀𝑖 ∈ Ω 𝑇𝐺
𝐏𝐀𝐓𝐇 𝐛𝐮𝐬 Matrix containing the vector 𝒑𝒂𝒕𝒉𝒃𝒖𝒔𝑖 with ∀𝑖 ∈ Ω𝑛𝑜𝑑𝑒
𝑇𝐺
𝒑𝒂𝒕𝒉𝒃𝒓𝑖 Vectors containing the branches in 𝑖 𝑡ℎ road
𝒑𝒂𝒕𝒉𝒃𝒖𝒔
𝑖 Vectors containing the buses in 𝑖 𝑡ℎ road
𝑃𝐶 𝑉𝐼 Penalty coefficient of voltage violation in $/kV
𝑃𝐹𝑖𝐷𝐷𝐺,1𝑠𝑡 Minimum operating power factor allowable of the 1st quadrant of the DDG unit
𝑖
𝑃𝐹𝑖𝐷𝐷𝐺,4𝑡ℎ Minimum operating power factor allowable of the 4th quadrant of the DDG unit
𝑖
xxvii
𝑃𝐹𝑖𝑇𝑅,1𝑠𝑡 Minimum operating power factor allowable of the 1st quadrant of the
transformer 𝑖
𝑃𝐹𝑖𝑇𝑅,4𝑡ℎ Minimum operating power factor allowable of the 4th quadrant of the
transformer 𝑖
𝑠𝑖𝑡𝑒 Vector containing the location of the PVs for sample 𝑠𝑝
𝑷𝑽𝑠𝑝
𝑷𝑽𝑑𝑜𝑤𝑛
𝑡
𝑑𝑜𝑤𝑛
Vector containing the status of the PVs (i.e., 𝑃𝑉𝑖,𝑡,𝑠𝑒 ) at time 𝑡
𝑄𝑖,𝑡,𝑠,𝑠𝑒 Reactive power injected to bus 𝑖 at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝐿
𝑄𝑖,𝑡,𝑠,𝑠𝑒 Reactive power load for bus 𝑖 at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝐿,𝑝𝑒𝑎𝑘
𝑄𝑖,𝑡,𝑠,𝑠𝑒 Original reactive peal load from data sheet, bus 𝑖 at time 𝑡 for 𝑠 𝑡ℎ scenario of
season 𝑠𝑒
𝑅𝑖𝐶 Certain solar radiation point of the ith PV unit, typically set to 0.15 kW/m2
𝑅𝑖𝑆𝑇𝐷 Solar radiation in standard test conditions of the 𝑖 𝑡ℎ PV un, typically specified
as 1000 W/m2
𝒓𝑑𝑛𝑛
𝑒𝑙𝑒,𝑠𝑝 Vector containing uniform random number between [0,1] for planning element
𝑒𝑙𝑒 for sample 𝑠𝑝
𝑟𝑖,𝑡,𝑠 𝑏𝑟
Uniform random number for generating 𝑎𝑖,𝑡,𝑠,𝑠𝑒
𝒓𝒎𝑠,𝑠𝑒
𝑚
Vector constraining 𝑟𝑡,𝑠,𝑠𝑒 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑚
𝑟𝑖,𝑡,𝑠,𝑠𝑒 Measured solar radiation of the 𝑖 𝑡ℎ PV unit at time 𝑡 for 𝑠 𝑡ℎ scenario of season
𝑠𝑒 (kW/m2)
𝑟𝑖𝐸𝑆 Ratio between the rated capacity and rated power for the ESS
𝒓𝒐𝒐𝒕𝑴𝑮 Vector containing root of each microgrids
𝑟𝑜𝑢𝑛𝑑(. ) Round deployment, “1” for input ≥ 0.5 and “0” otherwise
𝑆𝑏𝑎𝑠𝑒 Setting power base value
S𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 Capacity rating of the DDG unit 𝑖
𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 Capacity rate of the transformer
𝑆𝑂𝐶𝑖0 Initial state of charge of the ESS unit
𝑆𝑃𝐹0 /𝑆𝑃𝐹 Expected actual system performance
𝑠𝑡𝑛𝑖𝑠𝑤 Status of ISW 𝑛
,
𝑡𝑟 Time 𝑡 corresponding to first failure branch revered
𝑡𝑑 Time 𝑡 corresponding to the end of event
𝑡𝑒 Time 𝑡 corresponding to the beginning of impact or severe event
𝑇 𝑒𝑣𝑒𝑛𝑡 Time segment period of the event Phase I
𝑡𝑟 Time 𝑡 corresponding to the full system recovery
𝑏𝑟,𝑓𝑎𝑖𝑙
𝑡𝑖,𝑠,𝑠𝑒 Time corresponding to the moment when branch 𝑖 fail for 𝑠 𝑡ℎ scenario of season
𝑠𝑒
𝑏𝑟,𝑟𝑒𝑝
𝑡𝑖,𝑠,𝑠𝑒 Time t when branch recovered or repaired for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑑
𝑡𝑠,𝑠𝑒 Time 𝑡 𝑑 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑒
𝑡𝑠,𝑠𝑒 Time 𝑡 𝑒 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑻𝑮𝑏𝑢𝑠 Vector containing the set of target node 𝑇𝐺𝑖𝑏𝑢𝑠 with ∀𝑖 ∈ Ω𝑛𝑜𝑑𝑒
𝑇𝐺
𝑇𝐿 Project lifetime, typical set by 10 years
TTR Time to repair
xxviii
𝑇𝑇𝑅𝑖,𝑠,𝑠𝑒 Time to repair of the failure branch 𝑖 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝒃𝒓,𝒇𝒂𝒊𝒍 𝑏𝑟,𝑓𝑎𝑖𝑙
𝑻𝑻𝑹𝑠,𝑠𝑒 Vector TTR containing 𝑇𝑇𝑅𝑖,𝑠,𝑠𝑒
𝑉 𝑚𝑎𝑥 , 𝑉𝑚𝑖𝑛 Maximum/Minimum voltage violation of the bus
𝑣𝑖𝑐𝑖 ,𝑣𝑖𝑐𝑜 ,𝑣𝑖𝑟𝑎 Cut-in, cut-out, and the rated speed of the ith wind DG, respectively
𝒗𝒎𝑠,𝑠𝑒
𝑚
Vector constraining 𝑣𝑡,𝑠,𝑠𝑒 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑚
𝑣𝑖,𝑡,𝑠,𝑠𝑒 Measured wind speed of the 𝑖 𝑡ℎ wind DG at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑉𝑏𝑎𝑠𝑒 Setting voltage magnitude base value
𝑙𝑜𝑤 ℎ𝑖𝑔ℎ Specified low and higher bound for critical wind
𝑣𝑐𝑟𝑖 /𝑣𝑐𝑟𝑖
𝑠𝑝𝑒𝑐
𝑉𝐺𝐶 Nominal specified voltage during grid-connected mode
𝑉𝑖∗∗ Voltage reference for secondary control of the DDG at candidate bus 𝑖 operated
with droop/2nd
𝑉𝑖∗ Voltage reference for primary control of the 𝑖 𝑡ℎ DDG at candidate bus 𝑖 operated
with droop/2nd
𝑔𝑢𝑠𝑡
𝒗𝑠𝑒 Vector containing 3-Second wind gust in proportional to 𝒗𝑤𝑜𝑟𝑠𝑡
𝑠𝑒
𝐦
𝐯𝑠𝑒 Matrix constraining 𝒗𝒎𝑠,𝑠𝑒 for season se
𝒗𝑤𝑜𝑟𝑠𝑡
𝑠𝑒 Vector of wind scenario contains the word wind speed in m/s
𝐖𝑖 Matrix of weight for 𝑖 𝑡ℎ hidden layer of DNN architecture
𝑤𝑖𝑙𝑜𝑎𝑑 Weighing coefficient of loads
𝑠𝑖𝑡𝑒 Vector containing the location of the WTs for sample 𝑠𝑝
𝑾𝑻𝑠𝑝
𝑾𝑻𝑑𝑜𝑤𝑛
𝑡
𝑑𝑜𝑤𝑛
Vector containing the status of the WTs (i.e., 𝑊𝑇𝑖,𝑡,𝑠,𝑠𝑒 ) at time 𝑡
𝐛𝐮𝐬 𝐛𝐮𝐬
𝐘𝑠,𝑠𝑒 3D matrix containing the time couple admittance matrix 𝐘𝑡,𝑠,𝑠𝑒 at time 𝑡 for 𝑠 𝑡ℎ
scenario of season 𝑠𝑒
𝑍𝑖,𝑡,𝑠,𝑠𝑒 Time coupled impedance of branch 𝑖 at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑍𝑖0 Initial impedance value for branch 𝑖
𝒂𝒅𝒏𝒏
0,𝑠𝑝 Vector input sample 𝑠𝑝 for DNN architecture
𝒂𝒅𝒏𝒏
𝑖,𝑠𝑝 Vector of neuron activation obtained by 𝑖 𝑡ℎ hidden layer according to sample
𝑠𝑝
𝑏𝑟,𝑛𝑒𝑤 𝑏𝑟 𝑏𝑟,𝑟𝑒𝑝
𝑎𝑖,𝑡,𝑠,𝑠𝑒 Updated number for 𝑎𝑖,𝑡,𝑠,𝑠𝑒 at time 𝑡 = 𝑡𝑖,𝑠,𝑠𝑒
𝑏𝑟 Binary element represents the availability of branch 𝑖 at time 𝑡 of season 𝑠𝑒
𝑎𝑖,𝑡,𝑠,𝑠𝑒
𝑑𝑛𝑛 ′ Neuron activation obtained by the output- layer according to sample 𝑠𝑝
𝑎𝑠𝑝
𝐷𝐷𝐺
𝜂 Power efficiency of the DDG
𝜂𝑖𝐸𝑆𝑆 Efficiency of the energy storage
𝜇𝑐𝑜𝑛𝑑 , 𝜎 𝑐𝑜𝑛𝑑 Mean and Standard deviation of the normal distribution for conductor’s fragility
curve
𝜇 𝑝𝑜𝑙𝑒 , 𝜎 𝑝𝑜𝑙𝑒 Mean and Standard deviation of the normal distribution for pole’s fragility curve
𝜋 𝑏𝑟,𝑛𝑣𝑢𝑙 Vulnerability threshold for distribution branch
𝑏𝑟 𝑝𝑜𝑙𝑒𝑠
𝜋𝑖,𝑡,𝑠𝑒 Failure probability of the branch 𝑖 at time 𝑡 of season 𝑠𝑒, considering 𝜋𝑖,𝑡,𝑠𝑒 and
𝑐𝑜𝑛𝑑
𝜋𝑖,𝑡,𝑠𝑒
𝑐𝑜𝑛𝑑 Failure probability for a conductor of branch 𝑖 at time 𝑡 of season 𝑠𝑒
xxix
𝑃𝑜𝑙𝑒 Failure probability for single pole of branch 𝑖 at time 𝑡 of season 𝑠𝑒
𝑝𝑜𝑙𝑒𝑠 Failure probability considering all poles of branch 𝑖 at time 𝑡 of season 𝑠𝑒
𝝅𝒘𝒊𝒔𝒐
𝒔,𝒔𝒆 Vector of probability for both wind and solar scenario for 𝑠 𝑡ℎ scenario of season
𝑠𝑒 obtained by forward reduction technic
𝛑𝑠𝑒 𝑏𝑟
Matrix in 3D format containing the failure probability 𝜋𝑖,𝑡,𝑠𝑒 for season 𝑠𝑒
𝜎 𝑐𝑢𝑟𝑡 Cost of load curtailment in $/kWh
𝜎 𝑓𝑢𝑒𝑙 Fuel cost for DDG in $/kWh
𝐷𝐷𝐺
𝜎𝑂&𝑀 Cost $/kWh for O&M the DDG
𝐸𝑆𝑆
𝜎𝑂&𝑀 Cost $/kWh/year for O&M the ESS
𝐼𝑆𝑊
𝜎𝑂&𝑀 Cost $/kWh/year for O&M the ISW
𝑃𝑉
𝜎𝑂&𝑀 Cost $/kWh/year for O&M the PV
𝑊𝑇
𝜎𝑂&𝑀 Cost $/kWh/year for O&M the WT
𝑃𝐶𝐶
𝜎𝑡,𝑠,𝑠𝑒 Market energy price in #/kWh corresponding to time 𝑡 for 𝑠 𝑡ℎ scenario of season
𝑠𝑒
𝐸𝑆𝑆 Installation cost $/kWh of ESS at candidate bus 𝑖
𝜎𝐼𝑁𝑆
𝐸𝑆𝑆,𝐸 Investment cost $/kWh for energy density rating for ESS
𝜎𝐼𝑁𝑉
𝐸𝑆𝑆,𝑃 Investment cost $/kW for power rating for ESS
𝜎𝐼𝑁𝑉
𝐷𝐷𝐺
𝜎𝐼𝑁𝑉 Investment cost for DDG in $/kW
𝑃𝑉
𝜎𝐼𝑁𝑉 Investment cost for PV in $/kW
𝑆𝑆𝑊
𝜎𝐼𝑁𝑉 Investment cost in $ per allocated ISW
𝑊𝑇
𝜎𝐼𝑁𝑉 Investment cost for WT in $/kW
Sets of Variables:
𝑿𝑴𝑰𝑳𝑷 Vector containing the decision variables for the MILP problem formulation
𝑿𝑶𝑳𝑬𝑽 Vector containing the decision variables for the outer level problem
Continuous variables:
𝐸𝑆𝑆
𝐸𝑖,𝑡,𝑠,𝑠𝑒 Energy SOC of the ESS unit 𝑖 at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑏𝑟
𝑃𝑘,𝑡,𝑠,𝑠𝑒 Active power flowing across the transmission branch 𝑘 at time 𝑡 for 𝑠 𝑡ℎ scenario
of season 𝑠𝑒
𝑐ℎ
𝑃𝑖,𝑡,𝑠,𝑠𝑒 Active power charge by 𝑖 𝑡ℎ ESS unit at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑑𝑐ℎ
𝑃𝑖,𝑡,𝑠,𝑠𝑒 Active power discharge by 𝑖 𝑡ℎ ESS unit at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝐷𝐷𝐺
𝑃𝑖,𝑡,𝑠,𝑠𝑒 Active power generated by DDG 𝑖 at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑃𝐶𝐶
𝑃𝑖,𝑡,𝑠,𝑠𝑒 Active power imported from the main grid across the 𝑖 𝑡ℎ PCC of the MG at time
𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑃𝑉,𝑢𝑠𝑒
𝑃𝑖,𝑡,𝑠,𝑠𝑒 PV’s adjusting power output and used at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑊𝑇,𝑢𝑠𝑒
𝑃𝑖,𝑡,𝑠,𝑠𝑒 WT’s adjusting power output used at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑏𝑟
𝑄𝑘,𝑡,𝑠,𝑠𝑒 Reactive power flowing across the transmission branch 𝑘 at time 𝑡 for 𝑠 𝑡ℎ
scenario of season 𝑠𝑒
𝐷𝐷𝐺
𝑄𝑖,𝑡,𝑠,𝑠𝑒 Reactive power generated by the 𝑖 𝑡ℎ DDG unit at time 𝑡 for 𝑠 𝑡ℎ scenario of
season 𝑠𝑒
xxx
𝐸𝑆𝑆
𝑄𝑖,𝑡,𝑠,𝑠𝑒 Reactive power generate/receive of the 𝑖 𝑡ℎ ESS unit, at time 𝑡 for 𝑠 𝑡ℎ scenario
of season 𝑠𝑒
𝑃𝐶𝐶
𝑄𝑖,𝑡,𝑠,𝑠𝑒 Reactive power imported from the main grid across the 𝑖 𝑡ℎ PCC of the MG at
time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑣
𝑢𝑖,𝑡,𝑠,𝑠𝑒 Integrator output of secondary voltage restoration of 𝑖 𝑡ℎ DDG unit operating
with droop/2nd at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑉𝑖,𝑡,𝑠,𝑠𝑒 Voltage magnitude of the bus 𝑖 at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝛿𝑖,𝑡,𝑠,𝑠𝑒 Voltage magnitude angle of bus 𝑖 at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝛿𝑓𝑀𝐺
𝑏𝑢𝑠
,𝑡,𝑠,𝑠𝑒
𝑏𝑢𝑠
Phase angle of the frequency controllable bus 𝑓𝑖,𝑡 at time 𝑡 for 𝑠 𝑡ℎ scenario of
𝑖,𝑡
season 𝑠𝑒
𝑀𝐺
𝛿𝑡,𝑠,𝑠𝑒 Phase angle of the bus where the frequency control act at time 𝑡 for 𝑠 𝑡ℎ scenario
of season 𝑠𝑒
Binary variables:
𝑏𝑐𝑢𝑟𝑡
𝑖,𝑡,𝑠,𝑠𝑒 Binary variable for load curtailment (1: load shaded and 0: load fed) for bus 𝑖 at
𝑏𝑖𝐷𝐷𝐺,𝑠𝑖𝑧𝑒 Integer variable for DDG allocation at candidate bus 𝑖
𝑏𝐸𝑆𝑆
𝑖,𝑡,𝑠,𝑠𝑒 Binary variable indicates the status of 𝑖 𝑡ℎ ESS at time 𝑡 for 𝑠 𝑡ℎ scenario of
season 𝑠𝑒, i.e. 1: discharge and 0: charge
𝑏𝑖𝐸𝑆𝑃,𝑠𝑖𝑧𝑒 Integer variable for ESS allocation at candidate bus 𝑖
𝑏𝑖𝐼𝑆𝑊 Binary decision variable for ISW allocation at candidate branch 𝑖
𝑏𝑖𝑃𝑉,𝑠𝑖𝑧𝑒 Integer variable for PV allocation at candidate bus 𝑖
𝑏𝑖𝑊𝑇,𝑠𝑖𝑧𝑒 Integer variable for WT allocation at candidate bus 𝑖
Function and Metric:

𝐴𝐸𝑀𝐼𝑆 𝐷𝐷𝐺 Annual CO2 emission in kg/kWh/year according to DDGs generation
𝐴𝐸𝑀𝐼𝑆𝐷𝐷𝐺,𝑑𝑙 𝐴𝐸𝑀𝐼𝑆 𝐷𝐷𝐺 based on DNN model
𝐴𝐸𝑀𝐼𝑆 𝐺𝑅𝐼𝐷 Annual CO2 emission in kg/kWh/year according to power imported from main
grid
𝐴𝐸𝑀𝐼𝑆 𝐺𝑅𝐼𝐷,𝑑𝑙 𝐴𝐸𝑀𝐼𝑆 𝐺𝑅𝐼𝐷 based on DNN model
𝐴𝐸𝑀𝐼𝑆 𝑂𝐿𝐸𝑉 Total CO2 emission, use as an objective function for the outer level problem
𝐴𝐸𝑀𝐼𝑆 𝑂𝐿𝐸𝑉,𝑑𝑙 𝐴𝐸𝑀𝐼𝑆 𝑂𝐿𝐸𝑉 based on DNN model
𝐴𝑂𝑀𝐶 𝐷𝐷𝐺 Annual O&M cost for DDGs in $/kWh/year
𝐴𝑂𝑀𝐶𝐷𝐷𝐺,𝑑𝑙 𝐴𝑂𝑀𝐶 𝐷𝐷𝐺 based on DNN model
𝐴𝑂𝑀𝐶 𝐸𝑆𝑆 Annual O&M cost for ESSs in $/kWh/year
𝐴𝑂𝑀𝐶 𝐼𝑆𝑊 Annual O&M cost for ISWs in $/year
𝐴𝑂𝑀𝐶 𝑃𝑉 Annual O&M cost for PVs in $/kWh/year
𝐴𝑂𝑀𝐶 𝑊𝑇 Annual O&M cost for WTs in $/kWh/year
𝐶𝐷𝐷𝐺,𝑑𝑙 𝐶 𝐷𝐷𝐺 based on DNN model
𝐶𝑈𝑅𝑇
𝐶𝑡,𝑠,𝑠𝑒 Cost of energy curtailment during IS mode at time 𝑡 for 𝑠 𝑡ℎ scenario of season
𝑠𝑒
𝐶 𝐷𝐷𝐺 Total cost in $ for DDGs considering 𝑇𝐿
xxxi
𝐷𝐷𝐺
𝐶𝑡,𝑠,𝑠𝑒 Cost of natural gas consumption by the DDG units at time 𝑡 for 𝑠 𝑡ℎ scenario of
season 𝑠𝑒
𝐶 𝐸𝑆𝑆 Total cost in $ for ESSs considering 𝑇𝐿
𝐶 𝐼𝑆𝑊 Total cost in $ for ISWs considering 𝑇𝐿
𝑃𝐶𝐶
𝐶𝑡,𝑠,𝑠𝑒 Cost of energy imported from the substation across the PCC of the microgrid at
𝐶 𝑃𝑉 Total cost in $ for PVs considering 𝑇𝐿
𝑉𝐼
𝐶𝑡,𝑠,𝑠𝑒 Cost of voltage violation at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝑉𝐼,𝑀𝐺𝑠
𝐶𝑡,𝑠,𝑠𝑒 Cost of voltage violation considering MGs at time 𝑡 for 𝑠 𝑡ℎ scenario of season
𝑠𝑒
𝐶 𝑊𝑇 Total cost in $ for WTs considering 𝑇𝐿
𝐶𝑃𝐶𝐸𝑆𝑆,𝑃 Total capital cost for ESSs in $/kWh
𝐶𝑃𝐶 𝐷𝐷𝐺 Total capital cost for DDGs in $/kWh
𝐶𝑃𝐶 𝐼𝑆𝑊 Total capital cost for ISWs in $
𝐷𝐷𝐺 Daily CO2 emission in kg/kWh according to DDGs generation
𝐸𝑀𝐼𝑆𝑠,𝑠𝑒
𝐺𝑅𝐼𝐷 Daily CO2 emission in kg/kWh according to power imported from main grid
𝐸𝑀𝐼𝑆𝑠,𝑠𝑒
𝐼𝑁𝑆𝐶 𝐸𝑆𝑆 Installation cost in $ for ESSs
𝑀𝐼𝐿𝑃
𝑂𝐹𝑠,𝑠𝑒 Normalized objective function for MILP problem at time 𝑡 for 𝑠 𝑡ℎ scenario of
season 𝑠𝑒
𝑀𝐼𝐿𝑃,𝑀𝐺𝑠 Normalized objective function for MILP problem considering microgrid
𝑂𝐹𝑠,𝑠𝑒
formations at time 𝑡 for 𝑠 𝑡ℎ scenario of season 𝑠𝑒
𝐷𝐷𝐺
𝑂𝑀𝐶𝑠,𝑠𝑒 Pre-representation O&M cost for DDGs in kWh/day for 𝑠 𝑡ℎ scenario of season
𝑠𝑒
𝑅𝐸𝑆𝐼 Basic resilience index
𝑅𝐸𝑆𝐼 𝑂𝐿𝐸𝑉 Resilience index considering SPF progression phase I and II, use as an objective
function for the outer level problem.
𝑅𝐸𝑆𝐼 𝑂𝐿𝐸𝑉,𝑑𝑙 𝑅𝐸𝑆𝐼 𝑂𝐿𝐸𝑉 based on DNN model
𝑇𝐶 𝑂𝐿𝐸𝑉 Total cost in $, use as an objective function of the outer level
𝑇𝐶 𝑂𝐿𝐸𝑉,𝑑𝑙 𝑇𝐶 𝑂𝐿𝐸𝑉 based on DNN model
𝑇𝐺𝑠𝑝,𝑛 Target n for training DDN model corresponding to LOPF solution of sample 𝑠𝑝
xxxii
This Thesis is dedicated to my beloved
parents and brother for their endless
encouragement, support and love
xxxiii
Remerciements
Ma plus profonde gratitude va à mes superviseurs, Professeur Jérôme Cros et Professeur Morad
Abdelaziz qui m'ont encadré tout au long de cette thèse. Je tiens à remercier pour leur gentillesse,
leurs conseils et leur soutien, leur disponibilité permanente et pour les nombreux encouragements
qu'ils m'ont prodigués. Un grand merci à eux pour m’avoir accepté et donné l'opportunité de découvrir
et d'apprendre beaucoup de choses merveilleuses en tant que leur étudiant doctorat.
Je tiens à remercier les membres de l’équipe du Programme Canadien de Bourse de la Francophonie

ou PCBF pour leur accueil chaleureux depuis mon arrivée au Canada. Mes séjours ont été bien remplis
par les merveilleuses aventures, les expériences et c’est grâce à leur collaboration et leur soutien tout
au long de mon séjour. Cette thèse n'aurait pas pu être réalisée sans leur soutien financier.
Rien de plus que ma plus profonde gratitude et un grand merci à mes parents et à mon frère pour leur
compréhension, leur patience, leur soutien et leurs encouragements interminables tout au long de mes
études de doctorat.
Je tiens à exprimer ma reconnaissance à toutes les personnes et organisations qui m'ont offert un
soutien constant pendant toutes mes années d'études de doctorat.
A cet effet, je tiens à remercier les membres du jury d'avoir accepté d'assister à la soutenance de cette
thèse.
Enfin, aucun acte n'est plus noble qu'une démonstration de gratitude. Mes remerciements les plus
profonds vont au LEEPCI, Département de génie électrique et de génie informatique; Faculté des
sciences et de génie; et spécialement à l'Université Laval pour m'avoir ouvert la porte et m'avoir
permis d'améliorer mes connaissances et mes compétences en tant qu’étudiant doctorat de l'Université
Laval à Québec, Canada.
Youthanalack Vilaisarn,
Quebec, Canada
xxxiv
Foreword
The thesis consists of 5 chapters in addition to an introduction and a conclusion. First, the introduction
has been originally written by the author of this thesis Youthanalack Vilaisarn and is presented at the
beginning of the thesis. Then chapter 1 introduces the background information pertaining to this
thesis, while additionally including the state of the art and literature review pertaining to the work
presented in this thesis. The literature review in chapter 1 has been extracted from the submitted,
accepted or published version of the articles presented by chapters 2 to 5. Finally, the conclusion has
been originally written by the author for concluding the work and presenting the future perspective.
A general description for chapter 1 through 5 is presented as follows:
The Introduction introduces general information about active distribution systems and the advantage
of the microgrids application in power distribution. Then, the research motivations are provided as
well as the summary of research objectives and the thesis outline.
Chapter 1 introduces the necessary background for the thesis’s research including the microgrids
planning and the microgrid operation and control aspects. Additionally, this chapter includes the
literature survey pertaining to the topics addressed in this thesis; namely: 1) the islanded microgrid
power flow problem, 2) the optimal power flow of the microgrid system, and 3) the general microgrids
planning and the specific topic related to the resilience oriented microgrids planning.
Chapter 2 presents the proposed branch based 𝐙𝐛𝐮𝐬 power flow algorithm for the balanced and
unbalanced droop controlled islanded microgrids. This method solves the non-linear power flow
problem without linearization and is accurate enough for validating the linear power flow to be
developed in chapter 3. This part has been published with original contribution in the Elsevier Journal
(IJEPES). The major details and results in this chapter has been summited and published as the
following papers:
• Paper: [1] Y. Vilaisarn and M. Abdelaziz, “An inversion-free sparse Z power flow algorithm
for large-scale droop controlled islanded microgrids,” Int. J. Electr. Power Energy Syst., vol.
121, no. August 2019, p. 106048, Oct. 2020.
• Status: published with « doi.org/10.1016/j.ijepes.2020.106048 »
xxxv
• Author: Youthanalack Vilaisarn; contribution: proposed the islanded microgrid 𝐙𝐛𝐮𝐬 power
flow algorithm, implemented the model, performed the simulation, developed the case
studies and prepared this paper.
• Co-author: Morad Abdelaziz; contributions: supervised the development of the proposed

model and its implementation, revised the code implementation, provided important insights
on the simulations and the interpretation of the results. Aided in the preparation of the
manuscript.
Chapter 3 presents a linear power flow model for balanced and unbalanced droop islanded
microgrids. This work shows higher computational efficiency by developing a linearized power flow
model. The model is node based and is later integrated into the optimal operation model that will be
developed in chapter 4. This original contribution has been presented in the conference CIRED Berlin
2020 and published in IET Library as the following paper:
• Paper: [2] Y. Vilaisarn, M. Abdelaziz, and J. Cros, “Linear Three-Phase Power Flow Model
for Autonomous Microgrid Systems,” CIRED - Open Access Proc. J., vol. 2020, no. 1, pp.
50–53, 2021.
• Status: published with « doi:10.1049/oap-cired.2021.0019 »
• Author: Youthanalack Vilaisarn; contributions: proposed the linear power flow model,
implemented the model, performed the simulation, developed the case studies and prepared
this paper.

manuscript.
• Co-author: Jérôme Cros; contributions: supervised the development of the proposed model
and its implementation, revised the code implementation, provided important insights on the
simulations and the interpretation of the results. Aided in the preparation of the manuscript.
Chapter 4 provided a time coupled MILP formulation for determining the optimal operation of
balanced and unbalanced AC microgrids. The formulation considers the microgrids operation with
hierarchical control during both GC and IS mode. The energy storage and renewable resource (i.e.,
xxxvi
WT and PV) have been accounted for. A major result of this chapter is submitted and published in
IJEPES of Elsevier as the following paper:
• Paper: [3] Y. Vilaisarn, M. Moradzadeh, M. Abdelaziz, and J. Cros, “An MILP formulation
for the optimum operation of AC microgrids with hierarchical control,” Int. J. Electr. Power
Energy Syst., vol. 137, no. September 2021, p. 107674, May 2022.
• Status: published with « doi.org/10.1016/j.ijepes.2021.107674 »
• Author: Youthanalack Vilaisarn; contributions: proposed the optimal linear power flow
model, implemented the model, performed the simulation, developed the case studies and
prepared this paper.
• Co-author: Majid Moradzadeh; contributions: Reviewed the paper, polished the paper in
terms of structure and mathematics, proposed several case studies.

manuscript.
and it implementation, revised the code implementation, provided important insights on the
Chapter 5 presents the resilience oriented microgrid planning framework. The planning problem is
based on a two-level optimization with 1) the outer level been formulated as a multi-objective
optimization problem to allocate the distribution system planning elements, and 2) the inner level
dealing with the stochastic optimization operation solution using time coupled MILP model delivered
from chapter 4. This chapter has been submitted on IEEE Access and with the details as follows:
• Paper: A Deep Learning based Multi-Objective Optimization for the Planning of Resilience
Oriented Microgrids in Active Distribution System”. Accepted for publication in IEEE
Access, July 2022.
• Status: Accepted for publication in IEEE Access by 30-July-2022
xxxvii
• Author: Youthanalack Vilaisarn; contributions: proposed the planning framework,
implemented the planning framework, performed the simulation, developed the case studies
and prepared this paper.
• Co-author: Yuri R. Rodrigues; contributions: Reviewed the paper, polished the paper in
terms of structure and mathematics, proposed several case studies.

manuscript.
and its implementation, revised the code implementation, provided important insights on the
The Conclusion concludes the research in this thesis and summarizes the contributions. Afterwards,
future perspectives are provided for future work.
xxxviii
Thesis Original Contribution
The general contribution of this thesis can be summarized as follows:
• A branch-based inversion free Zbus power flow algorithm for droop islanded microgrid is
proposed. The proposed algorithm does not require bus renumbering, nor does it require a
DG to be allocated at the reference bus. Moreover, the proposed method is particularly well
suited for sparse implementation and delivers an enhanced computational efficiency in the
power flow solution, making it particularly useful for large scale microgrid systems
comprising up to thousands of buses.
• Proposed a node based LPF model that accounts for the droop-controlled autonomous
microgrid philosophy of operation as well as considering both balanced and unbalanced
network characteristics.
• Proposed a time coupled MINLP and MILP OPF models for the optimum operation problem
of balanced and unbalanced microgrid with hierarchical control structure. The proposed
models consider the microgrid philosophy of operation during GC and IS. Moreover, the
energy storage system and the renewable energy resources i.e., WTs and PVs have been
accounted for.
• Proposed a general framework using meta-heuristic multi-objective optimization for planning

resilience oriented microgrid(s) in active distribution systems considering economic,
resilience and environmental points of views. The proposed framework employs novel deep
learning models to speed up the determination of the economic, resilience and environmental
metrics pertaining to the MG(s).
xxxix
Introduction
0.1 Thesis motivation

Power distribution systems are currently undergoing a major transition towards reduced carbon
emissions and increased sustainable development. The key drivers for this transition are the need to
control electricity costs, mitigate global climate change, improve power system reliability and
resilience, replace aging infrastructure, and provide reliable electricity to new developing
communities [4]. One enabling solution to this transition is making good use of distributed generation
(DG) units, especially those based on solar and wind energies, different kind of energy storage
systems (EESs) and active load control strategies [5]. Based on this transition, power distribution
networks have been changing towards higher penetration of renewable energy recourses along with
increased application of energy storage technologies and are as such transforming from their
traditionally passive structure into an active distribution system (ADS). This transition towards the
ADS structure brings benefits to both utilities and customers, as well as respond to the key driving
techno-economic issues.
The ADS paradigm enhances the distribution system operators’ (DSOs) ability to operate and manage
the distribution of electricity by offering new possibilities and flexibility [6]. In this context,
microgrids (MGs) have been put forward as an attractive new concept. The microgrid is proposed as
the building block of future active distribution systems. The microgrid is an electrical region within
the active distribution system with the capability to operate in the grid connected where it is connected
to the upstream network; or in the islanded mode of operation where it is operated in isolation from
the upstream network. Here, clustering active distribution system into microgrids can offer significant
benefits for both the utility and the customers [7]. Here, they are some examples of such benefits:
• Support the grid performance: the utmost value of microgrids for DSOs is to support the grid
performance by: a) improving the reliability of power supply by enabling self-healing at the
local region; b) improving the resiliency which refers to the capability of the power system
to withstand a low probability high-impact event such as an extreme weather events, natural
disasters, cyber-attack…etc. c) Improving the power quality for the customers’ by providing
services such as voltage support.
1
• Reduction of carbon emissions: microgrids can enable wider penetration of renewable DG
units in the distribution system, leading accordingly to a significant reduction in the overall
system carbon footprint.
• Enhanced economics: the proximity of local generation/load of the microgrid can bring back
the investment benefit in term of reduced losses and the transmission and distribution (T&D)
payment as well as reduce the energy purchasing from the main.
• Enable Increased Demand Response: microgrids can facilitate the implementation of demand
side management (DMS) or demand respond. The DMS programs encourage the consumers
to modify the pattern of electricity usage to reduce their consumption payment instead of
planning more generation.
While the deployment of microgrids is rapidly growing, its widespread adoption is still facing many
challenges in aspects such as planning and the availability of steady state analysis tools/simulators to
accurately account for the microgrid operational characteristics. Additionally, many distribution
utilities worldwide are increasingly suffering from power outages due to rare but intense impact
events known as High-Impact Low Probability (HILP) events. While microgrids can help in
alleviating the impacts of these HILP events, there is still a need for a general framework for planning
microgrids targeted at alleviating such contingency events. One of the utmost challenges to address
in this regard is the need to incorporate an accurate microgrid model in such planning methodology.
Moreover, the uncertainty due to the presence of renewable energy resources, loads and the
contingencies introduce complexity in the planning problem and consequently affect the efficacy of
the planning solution. Regarding the described motivation and problematic, the challenges can be
summarized as follows:
• There is a lack of a framework for planning resilience oriented microgrids that consider the
microgrid optimal operation under the hierarchical control structure. Additionally, such
planning framework need to also account for the effect of the uncertainties arising from the
stochastic nature of the contingencies and those of the renewable resources.
• The microgrids are typically operated and controlled via a hierarchical control structure. Yet,
this control structure has not been accounted for in the optimal operation models existing in
the literature. The lack of this consideration can lead to infeasible and/or inaccurate solutions.
Moreover, the optimal operation model should be computationally efficient, allow for time
coupling and be able to capture the operation constraints for different kind of resources (i.e.,
2
DDG, WT, PV and storage). These considerations are needed to allow for the application of
the optimal operation model in different stochastic planning and real time applications.
• There is a need for a computationally efficient power flow model of the droop-controlled
islanded microgrids. While, the power flow models of droop-controlled islanded microgrids
have been heavily studied in the recent literature, still all of the previous work in this area
focused on the nonlinear power flow equations which can be computationally demanding to
solve. As such, there is still a need for a fast and accurate linearized power flow model of
droop-controlled islanded microgrid that accounts for the islanded microgrid operation and
control philosophy. A computationally efficient linearized droop-controlled islanded
microgrid model is crucial for many stochastic planning and real time optimal operation
problems.
• The typical power distribution system that are to be operated in islanded microgrid mode can
comprise up to thousands of buses. Yet, the islanded microgrid power flow analysis
algorithms found in the literature are not suited for such large-scale systems. This challenge
put forward a need for an accurate islanded microgrid power flow algorithm suited for large
scale systems comprising thousands of buses. The needed algorithm should not only account
for the operation and control philosophy of the droop controlled islanded microgrid but
should also be computationally efficient and accurate.
0.2 Research Objective

The ultimate objective of this research is to propose a general flamework for the planning of resilience
oriented microgrids in active distribution systems. To this end, this research also concentrated on the
development and deployment of the necessary power flow and optimal power flow models needed to
accurately account for the operation of the microgrid system within the planning framework. The
specific objective of this research can as such be summarized as follows:
• Objective 1: Developing a power flow algorithm suitable for large scale balanced and
unbalanced droop controlled islanded microgrids.
• Objective 2: Developing a node based linear power flow model for balanced and unbalanced
microgrid systems operated in droop controlled islanded mode.
3
• Objective 3: Developing a mixed integer linear programming model for the optimum
operation of the AC microgrids with hierarchical control.
• Objective 4: Proposing a general framework for resilience oriented microgrids planning in

active distribution networks.
This objective has been interlinked and can be shown in figure as follows:
A branch based power flow Objective 1:

algorithm for large scale systems Branch based Power Flow
Linear power flow model for Objective 2:

islanded microgrid Linear Node-based Power Flow
optimum operation for grid- Objective 3:

connected and islanded microgrids Optimal Power Flow
General planning framework for Objective 4:

resilience oriented microgrids Stochastic Planning Scheme
FIGURE 0. 1: Basic outline of the research objectives

0.3 Thesis Outline
First, Chapter 1 presents the necessary background and the state of the art for the research study.
Additionally, this chapter introduce the specific literature survey on the microgrids steady state
analysis and planning application.
Chapter 2 introduces the inversion free Zbus algorithm for the droop islanded microgrid power flow.
This algorithm is based on branch-based method and has been initially developed for balanced
systems and was then extended to unbalanced system while still considering the special characteristics
and philosophy of operation of droop controlled microgrids. The proposed algorithm has been tested,
validated and was shown to be superior in terms of computational efficiency. Moreover, the proposed
algorithm has been tested with the large-scale IMG and was shown to be able to handle the practical
systems comprising up to a thousand of buses.
Chapter 3 derivers the LPF model for droop controlled islanded microgrid. This model is later
integrated in the optimal operation application presented Chapter 4. The developed model accurately
accounts for the philosophy of the microgrid operation in droop control in either balanced or
unbalanced conditions. The accuracy and the effectiveness of the proposed LPF model has been tested
4
across several balanced and unbalanced test system, and the results were validated by comparison
with other methods proposed in the literature and with those of the proposed Zbus method introduced
in Chapter 2.
Chapter 4 starts by providing the time coupled MINLP model for the optimum operation of the
microgrid with hierarchical control. Subsequently, the LPF model from Chapter 3 is adopted to
account for the microgrid operational characteristics as linear constraints. Then, linearization is
performed for 1) the absolute value operation in the objective function, and the 2) quadratic
constraints. By using these linearization and the LPF model, the developed MILP OPF model
provides the basis for accounting for the microgrid optimal operation as needed in the microgrids
planning framework introduced in Chapter 5.
Chapter 5 introduces the ultimate goal of the research by providing the general framework for
resilience oriented microgrids planning for distribution system. The proposed framework delivers the
optimal allocation of the resources (i.e., DDGs, WTs, PVs and ESSs) and the ISWs, that
simultaneously improve the distribution system resiliency against N-K contingencies as well as
account for the economic and environment impacts viewpoints. The optimal resources and ISWs
allocation are based on the multi-objective optimization problem, that incorporates the stochastic time
coupled optimal operation model developed in Chapter 4. Then, a deep learning approach is adopted
for dealing with the computation burden arising from the stochastic consideration and the need for
many optimum operation simulations.
Finally, the last chapter presents the general conclusion including the thesis summary, contribution
and the specific direction for the future work.
5
Chapter 1 Research Background and State of Art
1.1 Introduction
In this chapter, the research background and state of the art are presented. First, the differences
between the conventional distribution system (the so-called passive distribution system) and the
active distribution system are presented. Then, we introduce the microgrids and their basic operation
and control characteristics. Subsequently, the background related to the research objectives, namely;
microgrids power flow, microgrids optimal operation (optimal power flow) and the microgrid
planning aspect are presented. Finally, the remainder of this chapter is dedicated to the literature
survey including 1) the survey of microgrids power flow algorithms 2) the survey of microgrids
optimal operation algorithms and 3) the survey on power system resilience enhancement related to
microgrids. Finally, a discussion of the literature survey is introduced leading the way to the proposed
research works.
1.2 Passive and Active Distribution System in the Power System

Conventional or passive distribution system (PDS) are being changed into active distribution systems
(ADS) due to the inclusion of DERs in the distribution level. The difference between PDS and ADS
can be defined by the direction of the power flowing in the distribution network [8]. In PDS, the
power flow is unidirectional where the power is flowing from main bus (substation) to the consumers
(loads). While, in the ADS bidirectional power flow is dominant, where local generators located at
given load points can feed other load points in the system. ADSs have local systems in place to control
different DERs (including the distributed generators (DGs) and energy storages) and loads, where
using these systems DSOs are able to manage/configurate the energy flow using flexible and effective
network topologies [6]. Generally, in active distribution systems, DGs include dispatchable and non-
dispatchable units. The former is a firm generator (e.g., natural gas micro-turbine and biomass
generator) that can be dispatched to deliver a constant amount of power. On the other hand, non-
dispatchable are stochastic generation units including wind turbines and photovoltaics. The power
production of these resources is stochastic and they cannot be dispatched along the planning horizon,
as their production is dependent on the stochastic wind speed and solar irradiance. The basic definition
and framework of ADS have been well acknowledged by different academic organizations, such as
CIRED [9] and IEEE [10]. Figure 1. 1 show the different structure of both systems.
6
G{1,2}
SW{1,2,3,4}
Smart meter {1→14}
11 12 13 14 SW3
DMS DER1
DER
11 12 13 14
2 3 4 5 6 7 SW1 SW2
1 1
Subsation 8 9 10 Subsation 2 3 4 5 6 7
SW3 DER2
DER
8 9 10
(a) (b)
Distributed Sectionalizing or Smart Communication link between
Load Bus/node DER
Energy Resource Isolate switch meter DMS and components
FIGURE 1. 1: Different features of typical PDS(a) and ADS(b)
A shown Figure 1. 1(a), a typical PDS consists of single or multiple feeders and several sectionalizing
switches. While Figure 1. 1(b) the ADS additionally the DERs. These devices are controlled via the
distribution management system (DMS) with the supervisory control. Traditionally, section switches
in PDS are always manually operated, whereas the operation in the ADS makes use of advanced
automation technologies.
1.2.1 Microgrids
The microgrid have been identified as a key component of the Smart-Grid. The microgrid is deployed
for improving the reliability and quality, increasing system energy efficiency, and providing the
possibility of grid-independency to individual local users. In [11] the DOE has given the definition
of the microgrid as: “a group of interconnected loads and distributed energy resources within clearly
defined electrical boundaries that acts as a single controllable entity with respect to the grid. A
microgrid can connect and disconnect from the grid to enable it to operate in both grid-connected or
island- mode.” The concept of the microgrid comprises micro sources which are typically smaller
generation units (<100 kW) and connected to the bus via the power electronic interface. Usually, the
microgrid(s) are located at the distribution level where the locally supplied demands (loads) are fed
at a low voltage (480 Volt or less) [12]. This kind of microgrid has the possibility of being connected
or disconnected from the main grid via the main isolation switch. This allows the microgrid to be
operated in the autonomous islanded mode, while balancing the generation and load within its
boundary. On the other hand, there are also the microgrids that are located in remote areas and that
have to interconnection to the main electrical grid. Microgrids typically employ a hybrid of renewable
7
energy, storage and dispatchable DGs in order to supply their customers without interruption.
Traditionally, the typical microgrid comprises the an interconnection between DERs (generation and
storage) and the different types of loads; and are controlled via a microgrid control center with smart
integrated devices [13]. Figure 1. 2 present a general scheme of a basic microgrid.
Control signals
Upstream DG ESS Downstream

main-grid Controller Controller main-grid
DG ESS
PCC 1MG PCC nMG
OR OR
DER DER
Grid
Grid Meter
Interface
Interface
Upstream Downstream nth
microgrid #1 Microgrid information measurements microgrid
Meter signals
Control signals
Microgrid Central
Controller
Distributed Distributed Energy Storage Sectionalizing Point of common

DER DG ESS PCC
Energy Resource Generation System or Isolate switch coupling
Power flow
Communication path Load Bus/node Smart meter
direction
FIGURE 1. 2: General scheme of the basic microgrid
First of all, it is easy to identify the electrical boundaries of the microgrids as the point of common
coupling (PCC1MG to PCCnMG this case). These points are interconnected with the external grid (another
microgrids or main-grid). It is possible to have different level of voltage between the microgrid and
the external grid. In this case the interface element in required and must be installed to permit the
interconnection between them. The control in microgrid can be centralized or decentralized [14] and
permits the operation as a single controllable entity. Thus, smart meters and smarts communications
infrastructure are typically required. Finally, the microgrid is permitted to operate in the islanded
mode by the switch at PCC. Depending on the technology of the microgrid infrastructure and the
devices installed, the microgrid can be operated in AC, DC of mixed. But in this research, we focus
on the planning of the microgrid in the AC active distribution system only.
8
1.2.2 Basic Droop Controlled Microgrid and the Implementation
As mentioned, the microgrid can be operated in grid-connected or islanded mode based on the grid
specific situation and the decisions of DSOs. In the grid-connected mode, the deficit in power between
the microgrid demand and generation can be supplied by the main grid. Similarly, excess power
generated in the microgrid can be traded with the main grid and used to provide ancillary services
[14]. Thus, the frequency is pre-specified, the voltage regulation are maintained at different microgrid
buses and given that the small size of the DG units compared to the power grid, the DG units within
the microgrids are controlled as PV or PQ buses [15]. However, in the islanded mode the DG units
are basically operated in droop mode to maintain balance of the generation and demand
simultaneously while controlling the voltage magnitude and frequency within the permissible ranges.
In this case, there is no slack bus which able to act as the reference in the network. On the other hand,
the microgrid can be operated with primary droop and the supervisory of secondary control
(droop/2nd). This kind of control structure guaranties that the microgrid is operated with the nominal
frequency and voltage as well as allows the microgrid be ready to re-synchronize to the main grid
when this is needed [16].
In the droop mode of operation, the power is shared with respect to the droop-characteristic. Here,
Figure 1. 3 present the steady-state, fundamental frequency model of the DG units operating in droop
mode [15]. In this model, the energy source, the power electronic converter and the output filter are
modeled as an ideal voltage source whose frequency and the voltage magnitude are controlled thought
the droop characteristics. This model is sufficient for the steady state analysis of droop controlled
islanded microgrid systems. Figure 1. 4 show the droop characteristic curves implemented by the
DG controller. In these figures, Figure 1. 4(a) illustrates the relation of the frequency and the active
power generated by the DG units; namely 𝑓/𝑃 curve. While, the Figure 1. 4(b) represents the reactive
power and voltage magnitude relations namely 𝑉/𝑄 curve.
PCC i DG
PiG,QiG
f/P and V/Q

Droop characteristic
control
|Vi|, f
GND
FIGURE 1. 3: Steady state, fundamental model of a DG unit operating in the droop mode
9
f(p.u.) V(p.u.)
f* f = f *- m pP G V* |V| = V * - n qQG
f m p low |V| n q low
m p high n q high
P(p.u.) Q(p.u.)
PG PG,max QG QG,max
(a) (b)
FIGURE 1. 4: Static Droop characteristics, 𝒇/𝑷 (a) and 𝑽/𝑸 (b)
Eq. (1.1) and Eq. (1.2) give droop characteristics implemented by the droop controller.
𝑓 = 𝑓 ∗ − 𝑚𝑝 𝑃𝐺 (1.1)
|𝑉̅ | = 𝑉 ∗ − 𝑛𝑞 𝑄 𝐺 (1.2)
In these equations above, 𝑓 and 𝑉 are the output frequency and voltage of the DG unit. While, 𝑓 ∗ and
𝑉 ∗ represented the frequency and voltage set point at no load. Given the per unit representation of the
power system, in 𝑓 is interchanged with 𝜔 in the literature. 𝑃𝐺 and 𝑄 𝐺 are the active and reactive
power generated by the DG unit. Finally, 𝑚𝑝 and 𝑛𝑞 expressed the static active and reactive droop
gain.
One way to calculate the static droop gains is to use the range of permissible frequency and voltage
deviation over the maximum capacity of generation of the DG units. Eq. (1.3) and (1.4) illustrate the
formulation for the droop gains calculation using this method:
𝑚𝑃 = (𝑓 𝑚𝑎𝑥 − 𝑓 𝑚𝑖𝑛 )⁄𝑃𝐺,𝑚𝑎𝑥 (1.3)
𝑛𝑞 = (𝑉 𝑚𝑎𝑥 − 𝑉 𝑚𝑖𝑛 )/𝑄 𝐺,𝑚𝑎𝑥 (1.4)
where, 𝑓 𝑚𝑎𝑥 and 𝑓 𝑚𝑖𝑛 represent per unit max and min allowable range of frequency in system. While,
𝑉 𝑚𝑎𝑥 and 𝑉 𝑚𝑖𝑛 are the per unit max and min permissible range for voltage. 𝑃𝐺,𝑚𝑎𝑥 and 𝑄 𝐺,𝑚𝑎𝑥
denoted the max and min capability of active and reactive power generated by DG units. Supposed,
𝑖 is the number of the droop-controlled DG unit in the microgrid; in case having a DG unit with
different rating; To achieve proper sharing of the load among these DG units, the static droop gains
should be calculated using the equations below [15], [17]:
10
𝑝 𝑝 𝑝
𝑚1 𝑃1𝐺,𝑚𝑎𝑥 = 𝑚2 𝑃2𝐺,𝑚𝑎𝑥 = ⋯ = 𝑚𝑖 𝑃𝑖𝐺,𝑚𝑎𝑥 (1.5)
𝑞 𝑞 𝑞
𝑛1 𝑄1𝐺,𝑚𝑎𝑥 = 𝑛2 𝑄2𝐺,𝑚𝑎𝑥 = ⋯ = 𝑛𝑖 𝑄𝑖𝐺,𝑚𝑎𝑥 (1.6)
The DG units in the microgrid are sharing active and the reactive power via droop control. Here, the
active power is shared through the global frequency of the system while the reactive power depends
on the output voltage of each DG unit. The Figure 1. 5 illustrates the droop sharing of the two DG
units in the microgrids.
f(p.u.) V(p.u.)
f* V*
|V1|
f |V2|
|Vi |
P(p.u.) Q(p.u.)
P 1G P 2G P iG Q 1G Q 2G Q iG
(a) (b)
FIGURE 1. 5: Droop sharing of two DG units in the microgrid
1.3 Conventional Power Flow and Introduction to Microgrid Power Flow
The power flow problem is the computation of the voltage magnitude and phase angle at each bus in
the power system at steady-state conditions. Based on these two variables, the real and reactive power
injected at each bus can be computed [18], [19]. Traditionally, given that the bus impedance matrix
𝐙̅𝐛𝐮𝐬 or the bus admittance matrix 𝐘
̅ 𝐛𝐮𝐬 are known, starting from equation (1.7) or (1.8) the power
flow problem can be solved:
̅𝑰 = 𝐘
̅ 𝐛𝐮𝐬 𝑽
̅ (1.7)
or by the inverse from,
𝑽 ̅ 𝐛𝐮𝐬 )−𝟏 𝑰̅ = 𝐙̅𝐛𝐮𝐬 𝑰̅

̅ = (𝐘 (1.8)
In the power flow analysis, each bus is associated with 4 quantities [20], namely the magnitude of
voltage (|𝑉̅ |), phase angle of voltage (𝛿), active power (𝑃) and reactive power (𝑄). For each bus, two
of these quantities are specified and the other two are required to be calculated through the power
flow solution. Figure 1. 6 represent the bus in the power system with their variables.
11
Generator
PiG From upstream bus
G V I = V i δi
QiG connected to bus i
Pi
Load Qi
PiL To downstream bus
Bus i
QiL connected to bus i
FIGURE 1. 6: Bus representation and its variables
As shown in the Figure 1. 6, they are four basic quantities associated in each bus 𝑘 including the bus
voltage magnitude 𝑉𝑖 , bus phase angle 𝛿𝑖 , real and reactive power injected to the bus 𝑃𝑖 and 𝑄𝑖 . While
the real and reactive power injected to the bus can be combined and obtained by:
𝑆𝑖̅ = 𝑃𝑖 + 𝑗𝑄𝑖 = (𝑃𝑖𝐺 − 𝑃𝑖𝐿 ) + 𝑗(𝑄𝑖𝐺 − 𝑄𝑖𝐿 ) ,∀𝑖 ∈ Ω𝐵𝑈𝑆 (1.9)
where, 𝑃𝑖𝐺 and 𝑄𝑖𝐺 are the real and reactive generated power flowing into the bus 𝑖; 𝑃𝑖𝐿 and 𝑄𝑖𝐿 are the
real and reactive load power flowing out from the bus 𝑖.
Conventionally, in the distribution system or microgrid operated in grid connected the buses are
classified depending on the two specified quantities as shown as Table 1. 1. Buses are divided into
three categories as shown in Table 1. 1.
TABLE 1. 1: Summary of bus classification and variables

Type of the bus Variables or quantities specified and un-specified of the bus
P Q V δ
Slack bus Unknown Unknown Known Known
Generator bus or PV bus Known Unknown Known Unknown
Load bus or PQ bus Known Known Unknown Unknown
A PQ bus the Active (P) and Reactive (Q) power demand at the bus are specified. While, for a
generator bus or known as PV bus P and V are specified. The magnitude of the voltage is kept constant
by adjust the field current (excitation system). Reactive power link to the quantity of adjust the field
current and possible to control the active power P due to governor system. Finally, slack bus or
infinity bus represent the infinity network in the power system which has a special generator bus
served as reference bus for the system. The slack can supply whatever power needed to make the flow
in system balanced due to any load change. Therefore, the slack bus can be assumed to have a fixed
voltage magnitude and its Phase angle is taken as a reference.
12
Accordingly, in the order to determine the un-specified variables, an iterative technique such newton
Raphson (NR)[20], backward-forward-sweep (BFS)[21], gauss seidel (GS) or Levenberg-Marquardt
[22], are deployed to determine the correct answer. This iterative technique is traditionally repeated
until the unknown quantities (an updating of the voltage at each bus per iteration for example) of the
system get closer to the correct answer. When those quantities no longer change from one iteration to
the next, the solution has converged to the correct answer.
On the other hand, in islanded mode, there is no slack bus in the system which can serve as reference
bus for the power flow calculation. Moreover, the active and reactive power generated from the DG
units are un-specified and might be governed by their droop characteristic [15], up to secondary
control [23]. Additionally, the frequency is not pre-specified i.e. 50/60 Hz and need to be determined.
According to this issues, the islanded power tools have been developed in the literature.
1.4 Optimum Operation

In the electrical power system, the optimum operation is a steady state analysis problem/tool
associated with higher level of system operation and control (e.g., DSO level). This tool is used for
identifying the optimal dispatch of generation resources, load and other controllable devices
according to some predetermined technical and economic considerations. The optimal operation
problem has been addressed based on the AC or DC optimal power (OPF) formulation [24].
Generally, the OPF problem can be formulated as the nonlinear and non-convex problem including
the objective function to be optimized and a set equality/inequality constraint which should be
satisfied to ensure the correct system operation. The OPF delivers the optimal operation state for the
system without violating system and equipment operation constraints [25]. The basic OPF problem
can be formulated as follows:
max 𝑜𝑟 min 𝑓 𝑂𝑃𝐹 (𝐱 𝐎𝐏𝐅 , 𝐮𝐎𝐏𝐅 ) (1.10)
subject to:
𝐠 𝐎𝐏𝐅 (𝐱 𝐎𝐏𝐅 , 𝐮𝐎𝐏𝐅 ) = 0 (1.11)
𝐡𝐎𝐏𝐅 (𝐱 𝐎𝐏𝐅 , 𝐮𝐎𝐏𝐅 ) ≤ 𝑜𝑟 ≥ 0 (1.12)
First, 𝑓 𝑂𝑃𝐹 (𝐱 𝐎𝐏𝐅 , 𝐮𝐎𝐏𝐅 ) in (1.10) is the objective function determined by the dependent variables
𝐱 𝐎𝐏𝐅 and control variables 𝐮𝐎𝐏𝐅 . Then, equation (1.11) is compact vectorial format represents the
13
equality constraints related to 𝐱 𝐎𝐏𝐅 and 𝐮𝐎𝐏𝐅 , while (1.12) is denoted for the inequality constraints.
In the power system, the objective function 𝑓 𝑂𝑃𝐹 (. ) can consider both economic and technical criteria
which is dependent on the specific application domain. Typically, 𝑓 𝑂𝑃𝐹 (. ) include the minimization
of generation cost, system losses, voltage deviation, etc. Then, the constraints represent the
characteristics of the system and their operation limit [26]. Typically, the equality constraints include
the power flow equations. The equality constraints also include particular models representing the
operation of specific devices and systems in the grid under consideration e.g., storage and microgrid.
On the other hand, the inequality constraint guaranties that the control variables 𝐮𝐎𝐏𝐅 obtained are
within some prespecified ranges e.g., generation and storage capacity. Additionally, the inequality
constraints limit the dependent variables 𝐱 𝐎𝐏𝐅 (e.g., voltage magnitude and system frequency) and
allow the system to be operated within the permissible and tolerable ranges.
The variables include the set 𝐱 𝐎𝐏𝐅 and 𝐮𝐎𝐏𝐅 which are dependent on the specific application.
Generally, 𝐱 𝐎𝐏𝐅 represented the solution state of the system e.g., frequency, voltage magnitude and
phase angle, while 𝐮𝐎𝐏𝐅 represent the optimum control variables of the optimization problem which
make the problem maximize or minimize its objective function. Typically, 𝐮𝐎𝐏𝐅 can comprise
variables of continuous nature (e.g., optimal generator power dispatch and the set point of generator
voltage controller), binary nature (e.g., generator unit commitment and load shedding decision), the
integer or a combination of all of them. The combination of different variables can lead the OPF
problem become nonlinear. In the optimum operation problem, one of the most important non-
linearity terms is the power equations. Moreover, the introduction of integer/binary variables renders
the problem Mixed Integer Non-Linear Programming (MINLP) problem. In this case, due to the non-
linearity or non-convex, the solution delivered by this OPF problem can be representative of local
optimal solutions. The OPF problem can be linearized, realizing better computational efficiency and
reaching the global optimal solution. The transformation from non-linear OPF problem to linear OPF
problem can be archived by the linearization of the non-linear term/equations of original non-linear
OPF problem [24][26]. Nonetheless, applying linearization to the problem can lead to loss of
accuracy. For this case, the validation process is required in the order to verify that any deviations are
within an acceptable tolerance.
1.5 Active Power Distribution System and Microgrid Planning

The planning of the distribution system is the problem intended to answer questions pertaining to how
to strategically meet the increase in load demand and to expand the electrical services driven by
different technical, economic, reliability, and environment aspects. Basically, the distribution system
14
planning today can be classified into the conversion planning and microgrid planning [8]. The first
type of planning focuses on the transformation from the PDS into the ADS by the integration of DERs
and storages in the power distribution network. In this case, the optimum allocation of the DER units
and distributed storage systems serves as a crucial solution to satisfy the increase in load demand. If
the DERs are integrated optimally, many benefits can be obtained such as reduced power losses,
improved voltage profile, enhanced reliability, enhanced economic and so on [6]. On the other hand,
microgrid planning is concentrated on the design of self-sufficient regions within the distribution
network. This can be achieved by deploying optimal microgrid(s) construction and can serve
difference perspectives e.g., self-sufficient, reliable and both [27] [28].
Generally, the ADS and MG planning can be categorized into the 1) investment planning, 2)
operational planning and 3) combination of both strategies. Additionally, the key features of the
planning problem can be categorized based on different perspectives such as the allocation problem
(e.g., DGs and storage placement), the operation problem and the uncertainty consideration problem.
An important aspect of the planning task is the planning objective. The planning objective represent
the goal of the planning which satisfy the stake holders involved (network operator and customers)
such as minimizing the costs, minimizing losses, improving reliability, minimizing carbon footprint,
etc. The objective of the planning can be a single objective, where in this case only one perspective
is considered. Or, it can be multi-objective where different objectives are combined together and
tradeoff between them are determined. To date, the planning problem formulated include the
numerous decision variables, more comprehensive objectives, complex constraints and with higher
level of uncertainties [6]. Here, the basic mathematical model of the planning is shown as:
max 𝑜𝑟 min 𝐹 𝑝𝑙𝑎𝑛 (𝐱 𝐩𝐥𝐚𝐧 ) = [𝑂𝐹𝑚 ]
𝐆 𝐩𝐥𝐚𝐧 (𝐱 𝐩𝐥𝐚𝐧 ) = 0 (1.13)

subject to {
𝐇 𝐩𝐥𝐚𝐧 (𝐱 𝐩𝐥𝐚𝐧 ) ≤ 𝑜𝑟 ≥ 0
Where, 𝐹 𝑝𝑙𝑎𝑛 (. ) denote the objective function representing the goal or the target of the optimization
problem. 𝐹 𝑝𝑙𝑎𝑛 (. ) can be a single objective or multi-objective and comprised 𝑂𝐹1 to 𝑂𝐹𝑚 .
Traditionally, the objective function can be the investment in the planning elements, operation and
maintenance costs or indices of reliability. Still, these are dependent on the perspective of planner.
Then, 𝐆 𝐩𝐥𝐚𝐧 (. ) and 𝐇 𝐩𝐥𝐚𝐧 (. ) are equality and inequality constraints, respectively, of the optimization
problem. Finally, 𝐱 𝐩𝐥𝐚𝐧 are the decision variables for the planning problem e.g., the location, size and
type of the DERs.
15
Nowadays, a major aspect of planning ADS or microgrids include the allocation of the intermittent
resource such as PV and wind. This mandates that the planning problem consider the uncertainty and
renders the planning problem more complex and harder to solve. Moreover, the increasing number of
faults arising due to natural disasters push forward the microgrids as a solution to increase reliability
and resilience. The microgrid concept has attracted the planners due to their ability to operate in grid
connected mode or islanded from the main grid. The microgrid planning can also coordinate the basic
conversion of PDS to ADS with the microgrid planning across using a multi-level optimization
framework that considers both investment and operation in the same framework [29]. Generally, the
outer level is made for the investment decisions pertaining to the planning elements. On the other
hand, the inner level deals with the microgrids operation strategy to hedge the difference stochastic
(e.g., renewable energy and contingencies) while respecting to the microgrids operation and control’s
behavior. This allows the planning problem able to capture the conversion from PDS to ADS as well
as considers the microgrid operation and control while considering the stochastic nature of resources
and possible faults in the system.
1.6 Literature Survey

1.6.1 Islanded Microgrids Power Flow
As discussed, the microgrid can be operated independently (i.e., islanded mode) or as a part of their
upstream network (i.e., grid connected mode). In the grid connected mode the microgrid operated
similar as an ADS, wherein the balance between the generation/demand were supported by their
upstream network and the DGs are controlled in a power-controlled mode. In this case, the traditional
power flow analysis is applicable. Nonetheless, in islanded mode the microgrid operated in the
absence of the infinite bus to impose the system frequency and system voltage magnitude.
Accordingly, the power flow analysis of islanded microgrids (IMG) exhibits several distinct features
compared to the traditional distribution system power flow problem; namely: (1) the absence of a
slack bus, (2) system frequency is one of the power flow variables, and (3) DG units are controlled
as droop units. These features have recently been recognized in the literature (e.g., [15], [30]). On the
other hand, practical distribution systems hosting IMGs can comprise thousands of electrical nodes
and there is nothing that preclude the possibility of creating an IMG comprising thousands of buses
in such systems. In these scenarios, despite comprising thousands of buses, the IMG cannot be treated
as a traditional distribution system due to the lack of an infinite bus in the system and requires an
IMG power flow algorithm.
16
Yet, existing IMG power flow algorithms in the literature are still not suited for the analysis of
practical large-scale IMG systems comprising thousands of buses. For instance, in [15] a node-based
IMG power flow analysis tool based on the Newton Trust Region (NTR) approach was proposed.
However, this algorithm is unsuitable for the sparse implementation required for a large-scale IMG
because this algorithm depends on matrix inversion processes which are generally complex to
perform for sparse matrices, due to the problem of fill-in and the large number of arithmetic
operations required (O(𝑁 3 )). The use of the famous Newton-Raphson (NR), and some of its variants,
was also investigated for solving the power flow problem for droop-controlled IMG systems in [31]–
[34]. Nonetheless, given their dependence on NR these algorithms, these algorithms are similar to
[15] in their dependence on matrix inversion processes. In [35] a branch-based algorithm was
proposed to solve the IMG power flow problem. Still, the algorithm in [35] depends on the incidence
matrix and its inversion, whose representation in sparse format will create the same problems
associated with sparse matrices inversion. Additionally, the algorithm in [35] requires that a DG unit
be located at the upstream bus (i.e., at the main grid bus) and that the voltage at the point of common
coupling (PCC) of this DG unit be used as a global variable to determine the reactive power
generation of the other DG units in the IMG. This leads to inaccurate representation of the V/Q droop
characteristics of the other DG units in the power flow analysis. To address the latter problem, the
work in [30] proposed the choice of the respective DG units PCC bus voltages as variables in order
to determine reactive power generation of the DG units accurately. However, the algorithm proposed
in [30] still requires that a DG unit be present at the upstream bus. A solution to handle this
requirement in the cases where no DG unit is not present at the upstream bus would be the
renumbering of the IMG buses. However, the renumbering of large-scale radial systems is generally
cumbersome and error-prone. In [36], a branch-based power flow algorithm for IMG that does not
require the presence of a DG unit at the upstream bus was proposed. However, this algorithm depends
on handling system elements/branches one by one and finding the shortest path between each node
and the reflection point which is problematic to do for large-scale systems, especially when the system
data is stored in sparse format. The work in [37] proposed a nested backward/forward sweep power
flow algorithm for droop-controlled IMG systems. Nonetheless, similar to [35], this algorithm
depends on the incidence matrix and its inversion. The algorithm in [38] proposed using conventional
power flow algorithms iteratively until the power flowing from the main grid is forced to zero.
However, this algorithm is dependent on running the power flow multiple times which is
computationally expensive especially for large-scale systems. Moreover, this algorithm also requires
a binary search for the voltage at the main grid bus, which add to the computational cost of the
algorithm.
17
On the other hand, the consideration of unbalance in the power flow problem of traditional
distribution networks has received much attention over the years. For instance, the work in [20]
proposed a direct approach for balanced and unbalanced load flow. In [39] and [40], an implicit Zbus
method incorporating a three phase power distribution model was used for an unbalanced load flow
application. Despite being suited for unbalanced distribution systems connected to the upstream
network, these works (e.g., [39]–[41]) do not consider the islanded operation of droop-controlled
IMGs and its implications on the power flow problem. Recently, the work in [42] and [43] proposed
unbalanced power flow methods for IMGs. Nevertheless, the methods proposed in [42] and [43]
mandate that DG units be located at the selected refence buses which creates similar problems to
those faced with [35]. In [44], the authors proposed an iterative branch-based method to compute the
power flow solution for balanced and unbalanced IMG systems. However, the algorithm in [44]
depends on the formation of the bus admittance matrix and its inversion to renew the network voltage
in each iteration, which is unsuitable for the sparse implementation required for a large-scale IMG
due to the problems of fill-ins and the large number of arithmetic operations required.
According to this review of literature, the existing power flow tools can be computationally
expensive. This issue renders the existing work not suitable for solving the power flow for a large-
scale IMG. Typically, the existing node-based methods, like Newton-based methods, need matrix
inversion processes which are generally complex to perform using the sparse matrices needed for
large scale IMG representation. On the other hand, while some branch-based IMG power flow are
proposed in the literature, most of them still require the incidence matrix and its inversion which will
create the same problem associated with the sparse matrices. Moreover, most of the existing branch-
based methods required a droop-controlled DG unit be located at the upstream bus which is require
bus renumbering to make them applicable to the general case. An alternative solution to deal with the
computation problem is using Linear Power Flow (LPF). Typically, the LPF models can provide fast
power flow solutions and enable a wide range of operational control and stochastic planning
algorithms that would otherwise be computationally infeasible [45]. Realizing the importance of such
models, different work in the literature have proposed balanced and unbalanced LPF models for the
distribution network (e.g., [46] and [47]). Yet, existing LPF models do not account for the droop-
controlled IMG special features in terms of the lack of a slack bus, the variability of autonomous
microgrid frequency and the operation of DG units in droop control mode. According to these
challenges, new power flow models/algorithms are required.
18
1.6.2 Microgrid Optimal Power Flow
In order to optimize and control the operation of the microgrid, the hierarchical control structure has
been widely adopted in the literature [48]. This structure employs several control layers, wherein each
of these layers has its own functionality and response time. A higher-level layer is typically employed
for economic purposes and manages the DG units and ESSs commitment and dispatching decisions
as well as the load curtailment decisions. Lower-level and intermediary-level layers are on the other
hand deployed for controlling the DG units power production and for maintaining the microgrid’s
voltage and frequency within bounds [49]. Still, the operation of the microgrid and its hierarchical
control structure are conceptually different in both grid-connected and islanding modes of operation.
In the grid-connected mode, the voltage and the frequency of the microgrid are imposed by the
upstream grid at the point of common coupling (PCC) and any mismatch between the microgrid’s
local generation and demand is supplied by the upstream grid. As such, in the grid-connected mode
of operation, the DG units are controlled to inject constant amounts of power prespecified by the
high-level layer of the hierarchical control structure [26]. On the other hand, in the islanding mode of
operation, the DG units cannot be controlled to supply some prespecified constant amount of power
and need to follow the load variations in order to ensure a stable system operation [14][50]. To this
end, in the islanding mode, primary droop control is deployed as the lower-level control layer in order
to control the DG units to follow the load variations. Secondary control is deployed as the
intermediary-level control layer responsible for compensating the steady-state errors in the system
voltage and frequency derived by the primary control. The secondary control hence ensures that the
microgrid’s voltage and frequency are maintained within bounds and that the microgrid is ready for
re-synchronization with the main grid when moving back from the islanding mode to the grid-
connected mode. Finally, tertiary control is adopted as the higher-level control layer used for ensuring
the economic viability of the microgrid operation by optimizing the microgrid’s commitment and
dispatch decisions.
The optimization of the operation of the microgrid resources and loads, performed by the higher-level
control layer of the microgrid’s hierarchal control structure, is based on an AC optimum power flow
(ACOPF) problem. In this context, the considered ACOPF problem seeks to optimize the microgrid’s
operation subject to the system’s operational constraints. Several research works have recently
attempted to accurately formulate and solve the ACOPF problem for microgrid systems. For instance,
a bi-level model has been proposed in [51] to optimally schedule the microgrid operation. However,
this model was restricted to the grid connected mode and the ability of the system to operate in
islanded mode was not considered. The mathematical formulation proposed in [52] has considered
19
the islanded mode of the microgrid. Yet, the primary and secondary control layers were not
considered, which can lead to infeasible solutions. In [53], an ACOPF for inverter based microgrids
operating in the islanded mode has been proposed. In this work, the primary droop control has been
considered and the microgrid operation was optimized for minimizing the distribution losses.
Additionally, an MINLP model for the optimal operation of unbalanced islanded droop based
microgrid has also been proposed in [26]. This formulation was also linearized and transferred into a
MILP model that can be solved with linear commercial solvers. While the work in [53], [26] has
overcome earlier drawbacks pertaining to the consideration of the primary droop control layer, still
these models were restricted to the islanded mode of operation and did not consider the secondary
control layer. Recently, a generalized model has been introduced in [54] to consider both the grid-
connected and islanded modes of operation of an unbalanced microgrid. However, the secondary
control layer has not been considered in this work. On the other hand, as the secondary control layer
change the DG units operating points and their active and reactive power injections, overlooking the
secondary control in the microgrid ACOPF problem formulation can lead to infeasible and suboptimal
operating points [16]. As mentioned, the existing works fall short in accounting the operation and
control of the microgrid with primary and supervisory with secondary control. In this sense, a new
ACOPF model for optimal operation the microgrid is required.
1.6.3 Power System Resilience Enhancement
The increasing recurrence of extreme weather events is demanding significant changes to the
distribution networks operation and design. Extreme weather events can cause power outages in what
would otherwise be a reliable distribution system. In fact, many distribution utilities worldwide are
increasingly suffering from power outages due to high-impact-low-probability (HILP) events. For
instance, in the ten years period from 2003 to 2013 more than 70 million customers in the United
States were affected by HILP events [55]. In order to address this issue, major efforts toward
distribution system resilience enhancement have been introduced recently. In this context, resilience
generally refers to the ability of the system to anticipate and withstand extreme HILP events, allowing
for a rapid recovery from possible interruptions due to the event [56]. To date, power distribution
utilities are in the lookout for practical solutions to increase their distribution systems’ resiliency,
allowing for continuous supply for critical loads, e.g., hospitals and data center, during HILP events
[57]. To this end, resilience enhancement solutions have been recently put forward in the literature.
Generally, these solutions can be categorized into 1) planning and infrastructure-oriented strategies,
and 2) operational strategies. The first kind of solutions mainly focuses on preventive actions to
protect the power distribution system against contingencies that may arise due to HILP events, e.g.,
20
DER allocation [58], optimal sizing [59] and hardening [60]. On the other hand, operational strategies
seek to improve the load restoration ability in case of a HILP event, e.g., pro-active operation [61],
network configuration [62] and MGs formation [63].
The concept of microgrids planning for resilience enhancement was recently introduced in the
literature. The majority of the work in this area opted for a multi-stage optimization problem for
delivering the optimal resilience enhancement solutions. Generally, the outer level is dedicated to the
investment decisions pertaining to the planning elements. On the other hand, the inner level deals
with the microgrids operation strategy to hedge the contingencies as well as to consider the microgrids
operation and control behavior. In [64], the authors proposed a two-stage model for planning
microgrids considering both economics and resilience enhancement. In this work, the determination
of the optimal site/size of the DERs and the backup distribution lines were considered. The problem
was formulated as a master/sub-problem and solved iteratively. The master problem focused on
deploying DERs and distribution lines while the sub problem identified the extreme events scenarios
allowing for up to N-3 contingencies. The solution delivered by this method improved the ability of
the microgrids to restore the power supply to loads during contingencies using either DERs or backup
distribution lines. Still, this work lacked in terms of accounting for the AC optimal power flow
representing the microgrid operation with hierarchical control. In [29] a stochastic planning model
was proposed for enhancing the distribution systems resilience using a microgrid formation strategy.
The proposed method was formulated as a multiple stage optimization model. The upper stage dealt
with the optimal line hardening and DG allocation while the lower stage dealt with the optimum
operation strategies (i.e., network reconfiguration, microgrids formation and demand side
management). This work considered the uncertainty of loads as well as the uncertainty of the damage
in the feeders caused by the weather events. Still, this work only considered dispatchable DG units.
Renewable energy resources and energy storage units were not considered. Additionally, the
microgrid hierarchical control structure was not considered in modeling the microgrid operation. The
work in [59] focused on developing a planning model for the optimal sizing of isolated networked
MGs considering the trade-off between economics and resilience enhancement. In this work, the
problem was formulated as a three level, wherein the first level focused on the optimal sizing of the
DERs and energy storage considering the normal operation (without contingencies). The second level
coordinated with third level allowed for considering both normal operation and emergency operation
of isolated networked MGs. For each of level, the AC optimal power flow has been used for delivering
the minimum of the generation cost, load shedding or both according to the operation condition (i.e.,
normal and emergency). A time coupled AC optimal power flow has been simulated for 4
21
representative days to allow for the uncertain characteristics of the seasonal load profile. Still,
renewable energy resources were not considered in this work. Moreover, the location of the DERs
and storage were pre-predetermined a priori to the solution. Authors in [65] proposed the optimal
sizing of mobile energy storage system in networked MGs. This work considered renewable energy
resources (i.e., PV), wherein different scenarios were adopted for considering the uncertain
characteristic of the PV generation as well as the load. Still, the microgrid hierarchical control
structure was not considered in modeling the microgrid operation. Moreover, this problem was
limited to the optimal sizing and placement of mobile energy storage, while the location and size of
PV and DGs were pre-predetermined a priori.
In regards to the existing resilience-oriented microgrids planning methodologies, one of the main
concerns is the computational burden. Attempting to manage their computational burden, existing
work endeavored to minimize the problem size by omitting renewable energy resources, pre-
specifying the location of some resources, neglecting the microgrid hierarchical control structure,
and/or considering a limited number of stochastic scenarios. While these techniques can reduce the
problem size, they on the other hand render the planning problem incomplete and only applicable to
particular planning instances. As a possible solution to this problem, the concept of deep learning can
provide an attractive solution for handling computationally demanding problems, i.e., its ability to
significantly increase the computation performance of large-scale analysis/optimization problems
make their solution more attainable in comparison to traditional methods. For instance, the concept
of deep learning has been applied for in the development of power flow and optimal power flow tools,
respectively, [66] and [67]. In these works, the deep learning technique was used for enhancing the
performance and computation efficiency of the power flow/optimal power flow solution approach.
Still, the concept of AC microgrids considering hierarchical control solutions has not been introduced
in this work and making its application not appropriate for microgrid planning problems. On the other
hand, the work in [68], proposes the use of deep learning for the restoration problem of critical loads
after experiencing a major outage. In this work, the deep Q-learning method is applied for speeding
up the solution approach, making the problem solvable for small distribution system up to large-scale
network. However, there is no consideration of the microgrid concept and its planning in this work.
Next, a new planning problem based on deep learning for a resilient transmission system is introduced
in [69]. In this work, a deep learning technique is adopted to forecast the load growth while improving
the forecasting task compared to an existing method. Still, the concept of microgrids is also not
introduced in this work. Recently, in [70], the concept of microgrid has been introduced while
coordinating several renewable resources and Plug-In Hybrid Electric Vehicles (PHEVs). In this
22
work, the deep learning technique is adopted to forecast the renewable output seeking to mitigate the
uncertainty caused by the renewable energy elements in the microgrid. While the deep learning
technique is adopted in this work, this work only discusses the operation stage and presents a
simplified problem by omitting the hierarchical control philosophy of the microgrid in its problem
formulation. Considering these challenges, a new framework for planning resilience oriented
microgrid(s) is need and developed in this thesis.
1.6.4 Discussion of the Literature Survey
According to the literature survey, steady state analysis, optimal operation and planning frameworks
that emphasize the microgrid concept have been heavily studied in the literature. Nonetheless, the
critical review of literature presented in this chapter shows that there are still several key challenges
that need to be addressed. First, the literature survey shows that the existing power flow tools are
mainly suited for small microgrid systems and cannot scale to large scale microgrid systems
comprising thousands of buses. This problem limits the existing power flow tools and make them not
suitable for many stochastic planning and real time operational problems. Thus, new power flow
models/algorithms are required. Furthermore, existing work in the area of ACOPF for the microgrid
systems fall short in accounting the operation and control of the microgrid with hierarchical control
(i.e., primary droop and secondary control). This can in turn lead to infeasible and sub-optimal
solutions. Thus, a new ACOPF model for optimal operation of the microgrid considering the
hierarchical control is required. Finally, the survey of literature in this chapter addressed the ultimate
challenges facing the resilience oriented microgrid planning and show that there is a need to consider
the operation of microgrids in the planning problem. Moreover, the survey shows that there is a need
of new methodology to deal the computation problem in the planning.
23
Chapter 2 Power Flow Algorithm for Islanded MGs
Y. Vilaisarn and M. Abdelaziz, “An inversion-free sparse Z power flow algorithm for large-scale
droop controlled islanded microgrid”. Published in International Journal of Electrical Power Energy
Systems, October 2020.
Résumé
L'une des principales caractéristiques des systèmes de micro-réseaux est leur capacité à fonctionner
de manière isolée du réseau principal. L’étude d’écoulement de puissance (EP) des micro-réseaux
isolé présente plusieurs caractéristiques distinctes par rapport à celui des méthodes traditionnelles,
comme : (1) l'absence le bus infini, (2) la fréquence est une variable à déterminer, et (3) les générateurs
sont contrôlés par le mode « statisme ». Plusieurs études dans la littérature ont traité ces particularités.
Cependant, les algorithmes existants qui servent à l’étude d’EP d’un micro-réseau isolé ne sont pas
adaptés à l'analyse un micro-réseau isolé à grande échelle qui se compose des milliers de nœuds
électriques. D'autre part, dans les systèmes de distribution pratiques, les micro-réseaux isolés créés
dans ces systèmes peuvent comprendre des milliers nœuds électriques. Ainsi, la mise en œuvre du
concept de micro-réseau dans les systèmes de distribution réels nécessite des outils l’étude d’EP
adaptés à l'analyse des micro-réseaux isolés à grande échelle. Pour cette raison, ce travail propose un
algorithme basé sur une matrice Zbus pour faire une étude d’écoulement de puissance d’un micro-
réseau isolé. Cet algorithme proposé est particulièrement adapté à une mise en œuvre une matrice
éparse ou creuse. Ceci permet de faire l’étude d’EP d’un micro-réseau qui se compose des milliers de
nœuds électriques. La précision et l'efficacité de l'algorithme proposé sont validées par plusieurs
études de cas portant sur des systèmes équilibrés et déséquilibrés.
Abstract
A main feature of microgrid systems is their ability to operate in isolation from the main grid, i.e.,
islanded microgrid (IMG) operation. The power flow (PF) problem of IMGs has been shown to
exhibit several distinct features compared to that of traditional distribution systems; namely: (1)
having no slack bus, (2) frequency being a PF variable, and (3) generators controlled as droop units.
Several studies in the literature have focused on and addressed these particularities. Still, existing
IMG PF algorithms are not suitable for the analysis of large- scale IMG systems comprising thousands
of buses. On the other hand, practical distribution systems and the IMGs created within such systems
may in many cases comprise thousands of electric nodes. As such, the implementation of the
microgrid concept in real distribution systems requires PF tools suitable for the analysis of large-scale
24
IMGs. This chapter proposes a Zbus PF algorithm for droop-controlled IMG that is particularly suited
for sparse implementation. The proposed algorithm is inversion-free, does not require bus
renumbering and is independent of third party sparse linear algebra libraries. The accuracy and
effectiveness of the proposed algorithm is validated through several case studies considering balanced
and unbalanced test systems.
2.1 Introduction
In this chapter we propose a Zbus power flow algorithm for droop-controlled IMGs of larger
dimensions. The proposed algorithm is inversion free and is particularly well-suited for sparse
implementation. The remainder of this chapter is organized as follows; First, Section 2.2 introduce
the proposed algorithm for balanced droop-controlled IMG. Subsequently, Section 2.3 discuss the
extension of the proposed algorithm to allow for the consideration of unbalanced IMGs as well as to
capture load voltage and frequency dependencies. The validation of the proposed algorithm is
discussed in Section 2.4, where balanced and unbalanced case studies are used to illustrate the
effectiveness of the algorithm. Finally, Section 2.5 summarize and conclude the chapter.
2.2 Proposed Algorithm

In droop-controlled IMGs, droop regulation is deployed to control the DG units output and achieve
an equilibrium between the demand and generation in the IMG. Accordingly, irrespective of the
generator type (i.e., whether machine based or inverter based) and its control, the DG units are
represented by their droop characteristics in the power flow problem [1]. In the droop mode, the
frequency and the voltage at the DG unit PCC depend on the DG units’ droop characteristic [15].
With the vector “𝕯” denoting the set of droop buses, the droop characteristic equations, 𝑓/𝑃 and
𝑉/𝑄, of the DG units in the IMG can be given as follows:
𝑝
𝑓 = 𝑓𝑖∗ − 𝑚𝑖 𝑃𝑖𝐺 ,∀𝑖 ∈ 𝕯 (2.1)
𝑞
𝑉𝑖 = 𝑉𝑖∗ − 𝑛𝑖 𝑄 𝐺𝑖 ,∀𝑖 ∈ 𝕯 (2.2)
where 𝑓 is the IMG’s frequency, 𝑓𝑖∗ and 𝑉𝑖∗ denote the no-load nominal frequency and voltage
magnitude set points of droop bus 𝑖, respectively. 𝑉𝑖 is the voltage magnitude at the PCC of the DG
𝑝 𝑞
unit connected to bus i. The active and reactive static droop gains are denoted by 𝑚𝑖 and 𝑛𝑖
respectively. 𝑃𝑖𝐺 and 𝑄𝑖𝐺 are the active and reactive power generated by the DG unit at bus 𝑖,
respectively. In the order to enhance the rate of convergence, the initial values for the active and
25
reactive power generation are initialized by distributing the total demand (neglecting line losses)
among the DG units in proportion to their droop gains.
In this work the Compressed Column Storage (CCS) format is used to store all matrices. Figure 2. 1
shows a flowchart of the proposed algorithm.
START
Initialization
Get PiL and QiL from datasheet,
|=1, δ =0, V1=1 0, f = f0 =1.
bus Pre-Initialize
Formation using (2.4)
Initialize droop bus power generation

PiG and QiG
Calculate = ( )*
bus
Calculate V1 +
Calculate mismatch ΔV Zbus loop
No
ΔV < tol
Yes
Calculate DGs branch current I bri

using procedure 1
Calculate frequency " f " using (2.10)
Update with new PiG and QiG

Outer loop
Calculate upstream bus voltage
V1 using procedure 2
Update Xbus using (2.14)
No
ΔV1 < tol
Yes
END
FIGURE 2. 1: Flowchart of the proposed algorithm
First, the sparse Zbus matrix is calculated and stored in rectangular form to enable updating the
imaginary part (which is frequency dependent) each iteration:
𝐙̅ 𝐛𝐮𝐬 = 𝐊 𝐓 𝐙̅𝐛𝐫 𝐊 (2.3)
𝐙̅ 𝐛𝐮𝐬 = 𝐑𝐛𝐮𝐬 + 𝑗𝐗 𝐛𝐮𝐬 (2.4)
where 𝐊 and 𝐙̅𝐛𝐫 are the incidence matrix and the primitive impedance matrix, respectively [71]. 𝐑𝐛𝐮𝐬
and 𝐗 𝐛𝐮𝐬 are a real and imaginary part of 𝐙̅ 𝐛𝐮𝐬. The proposed algorithm then uses two nested loops.
26
In the inner loop, an implicit Zbus method is implemented. The Zbus method run the power flow every
̅ and the voltage deviation vector ̅̅̅̅
iteration (𝑖𝑡) in order to update the bus voltage vector 𝑽 𝒅𝑽 = 𝐙̅𝐛𝐮𝐬 ̅𝑰.
By using the superposition, the voltage is updated until the mismatch 𝛥𝑉 = 𝑚𝑎𝑥(|𝑽𝑖𝑡 – 𝑽𝑖𝑡−1 |) is
less than a given tolerance. The formulations are given directly in Figure 2. 1.
In the outer loop, the frequency and reference bus voltage are updated. In this work, similar to [35]
and [30], the upstream bus #1 is chosen as the system reference. However, unlike [35] and [30], in
this work a DG unit may or may not be located at bus #1. Here it is worth noting that in [35] and [30],
𝑓 and 𝑉1 were updated using the branch power of the branch adjacent to the upstream bus (denoted in
̅ ). However, when no DG is located at the upstream bus, 𝑉1 and 𝑓 cannot be updated
this work by 𝐼1𝑏𝑟
this way. Accordingly, in this work procedures 1 and 2 are used to update 𝑓 and 𝑉1, as shown in
Figure 2. 1, based on the DGs branch currents.
Procedure 1: Calculate the branch current element 𝐼𝑖𝑏𝑟

̅ of ̅𝑰𝒃𝒓
Step 1: Calculate the complex branch current vector based on the injected current and the incidence
matrix
̅𝑰𝒃𝒓 = 𝐊𝑰̅ (2.5)
Step 2: Calculate the sum of branch current adjacent to bus 1
̅ = ∑𝑖∈𝐴𝑑𝑗𝑎𝑐𝑒𝑛𝑡 𝑡𝑜 𝑏𝑢𝑠 1 𝐼𝑖𝑏𝑟

𝐼1𝑏𝑟 ̅ (2.6)
Step 3: Based on (2.5) and with 𝐼𝑖 being 𝑖 𝑡ℎ element of 𝑰, update 𝐼𝑖𝑏𝑟
𝐼𝑖𝑏𝑟
̅ = 𝐼1𝑏𝑟
̅ − (−𝐼1𝑏𝑟
̅ − 𝐼𝑖̅ ) ,𝑖 = 𝕯(1) (2.7)
𝑏𝑟
̅ 𝑏𝑟
̅
𝐼(𝑖+1) = −𝐼(𝑖+1) ,∀𝑖 ∈ 𝕯 (2.8)
From the new 𝐼𝑖𝑏𝑟

̅ , the frequency can be directly calculated by (2.9) and (2.10). Denote by 𝑉̅𝑖 the 𝑖 𝑡ℎ
̅ from the inner loop and the frequency expression becomes:
element of vector 𝑽
𝑝 ∗
𝑓𝑖𝑥 = 𝑓𝑖∗ − 𝑚𝑖 [𝑃𝑖𝐿 + 𝑟𝑒𝑎𝑙(𝑉̅𝑖 (𝐼𝑖𝑏𝑟
̅ ) )] ,∀𝑖 ∈ 𝕯 (2.9)
27
𝑓 = [∑𝑖=𝐷𝐺 𝑓𝑖𝑥 ]/𝑛𝐷𝐺 ,∀𝑖 ∈ 𝕯 (2.10)
where 𝑓𝑖𝑥 is the pre-updated frequency considering a droop bus 𝑖. The pre-updated frequency 𝑓𝑖𝑥 in
(2.9) is calculated based on the equilibrium condition for active power in each droop bus node (i.e.,
̅ )∗ ) used in equation (2.1). As such, 𝑓𝑖𝑥 in (2.9) depend on “𝑃𝑖𝐿 +
𝑃𝑖𝐺 = 𝑃𝑖𝐿 + 𝑟𝑒𝑎𝑙(𝑉̅𝑖 (𝐼𝑖𝑏𝑟
∗
𝑟𝑒𝑎𝑙(𝑉̅𝑖 (𝐼𝑖𝑏𝑟 ) )” which represent the summation of the active power demand and the active branch
power flowing out from the 𝑖 𝑡ℎ bus where the DG is located. The new frequency is computed, from
(2.10), and is then used to update 𝑃𝑖𝐺 using (2.1). Afterward, 𝑄𝑖𝐺 can be updated directly using (2.2)
with the updated voltage 𝑽 from the inner loop. Then, the procedure 2 is used to determine 𝑉1 as
follows:
Procedure 2: Calculate reference bus voltage 𝑉1
Step 1: Calculate the magnitude of the new DGs voltage from DGs branch current obtained by (7)
and (8)
∗
‖𝑉𝑖𝑥 ‖ = 𝑉𝑖∗ − 𝑛𝑖𝑞 [𝑄𝑖𝐿 + 𝑖𝑚𝑎𝑔 (𝑉
̅𝑖 (𝐼𝑏𝑟
̅𝑖 ) )] ,∀𝑖 ∈ 𝕯 (2.11)
̅ from the inner loop with new ‖𝑉𝑖𝑥 ‖ and 𝛿𝑖 from the inner loop
Step 2: Update 𝑉̅𝑖 of vector 𝑽
𝑉̅𝑖 = ‖𝑉𝑖𝑥 ‖sin (𝛿𝑖 ) + 𝑗‖𝑉𝑖𝑥 ‖cos (𝛿𝑖 ) ,∀𝑖 ∈ 𝕯 (2.12)
Step 3: Update the element of ̅𝑰 with new 𝑉̅𝑖 , 𝑃𝑖𝐺 and 𝑄𝑖𝐺 . And then recalculate new ̅̅̅̅
𝒅𝑽.
̅ and ̅̅̅̅
Step 4: 𝑉1 can be calculated using the number of droop buses and the two vectors 𝑽 𝒅𝑽.
̅̅̅̅𝑖 ||] /𝑛𝐷𝐺

𝑉1 = [∑ ||𝑉̅𝑖 − 𝑑𝑉 ,∀𝑖 ∈ 𝕯 (2.13)
Similar to (2.9), in equation (2.11) the voltage magnitude is updated through the reactive power
generation equilibrium and respecting (2.2). To note here that the use of the variables and matrix with
the superscript “𝑥” is limited to the scope of procedure 2 only and does not affect processes outside
procedure 2. Also, the bar over the variables, vectors and matrices represents the complex
representation. According to the calculated frequency, the 𝐙̅𝐛𝐮𝐬 impedance is then updated. As
28
intended from the beginning, only 𝐗 𝐛𝐮𝐬 will be updated and this alleviate the need to rebuild 𝐙̅ 𝐛𝐮𝐬.
The update expression is given as follow:
𝑓
𝐗 𝐛𝐮𝐬 = 𝐗 𝐛𝐮𝐬,(𝟎) (2.14)
𝑓0
Where 𝑓 0 denote the system frequency of nominal operating point, and in this work 𝑓 0 set to one.
The algorithm ends when the mismatch 𝛥𝑉1 = |𝑉1𝑖𝑡 − 𝑉1𝑖𝑡−1 | of the outer loop is less than a
prespecified tolerance.
2.3 Extended to Unbalanced Systems

The proposed algorithm can be extended to allow for the consideration of three-phase unbalanced
systems. Figure 2. 2 shows a flowchart describing the implementation of the proposed algorithm in
three-phase unbalanced systems.
In the initialization stage the three-phase bus impedance matrix 𝐙̅ 𝐛𝐮𝐬,𝟑Ф is formulated in rectangular
form to enable updating the imaginary part as the iterations progress. The formulation of 𝐙̅𝐛𝐮𝐬,𝟑Ф for
unbalanced systems can be given as follows:
𝐙̅ 𝐛𝐮𝐬,𝟑Ф = (𝐊 𝟑Ф )T 𝐙̅𝐛𝐫,𝟑Ф 𝐊 𝟑Ф (2.15)
where 𝐊 𝟑Ф and 𝐙̅𝐛𝐫,𝟑Ф are the incidence matrix and primitive impedance matrix of the unbalanced
IMG, respectively. More details about formulating 𝐙̅𝐛𝐮𝐬,𝟑Ф can be found in [39]. To note here, that
𝐙̅𝐛𝐮𝐬,𝟑Ф , 𝐊 𝟑Ф and 𝐙̅𝐛𝐫,𝟑Ф are stored in CCS format similar to their corresponding matrices in the
balanced case. The initial values for the active and reactive power generation are initialized by
distributing the total demand (neglecting line losses) among the DG units in proportion to the DG
units respective droop gains considering the respective phase loading, where the three phases are
denoted by the superscripts a, b and c and the detailed procedures for initializing the DGs generation
will be provided later in this section. The initialization of the reference bus voltage is realized by
using procedure 4, which will also be explained later in this section. The proposed algorithm then
uses three nested loops. The inner and outer loops are similar to the ones used in the balanced case,
where in the inner loop an implicit Zbus method is implemented and in the outer loop the frequency
and reference bus voltage are updated. The middle loop is dedicated to calculating the power
distribution among the three phases. Procedure 3 describes how the DGs’ three-phase branch power
are calculated.
29
START
Initialization
Get P L,abc and Q L,abc from datasheet
abc
=[1 1 1]T, δabc =[0 -120 120]T,
f 3Φ = f0 =1
bus,3Ф
Formation using (2.15)
Pre-Initialize
Initialize generation power
PiG,(a,b,c) and QiG,(a,b,c)
Initialize 3Ф using Procedure 4 with

(QiL,3Ф +Qibr,3Ф)=QiG,3Ф+QiL,3Ф
abc abc abc *

=( / )
abc Ф bus,3Ф abc
= + no Zbus loop
ΔV<tol
yes
Calculate DGs branch power
Pibr,3Ф and Qibr,3Ф using Procedure 3
Update system frequency f 3Ф using (2.22)

br,3Ф
Update and Xbus,3Ф
Update PiG,3Ф and QiG,3Ф using (2.24) and (2.25)
Calculate PiG,(a,b,c) and QiG,(a,b,c)

no using (2.30) and (2.31)
abc
=(PG,abc-PL,abc)+j(QG,abc-QL,abc)
abc
=( abc/ abc )*
Outer loop
abc
= 3Ф + bus,3Φ abc Middle loop
no
Calculate mismatch ΔVmid
ΔVmid<Tol
yes
Update load demand PiL,(a,b,c) and QiL,(a,b,c)
using (2.42) and (2.43)
Update vector of upstream bus

voltage abc using Procedure 4
ΔV1<Tol
yes
END
FIGURE 2. 2: Flowchart of the proposed algorithm for unbalanced system
30
Procedure 3: Calculate the three-phase active and reactive branch power 𝑃𝑖𝑏𝑟,3Ф and 𝑄𝑖𝑏𝑟,3Ф
Step 1: Calculate branch current vector of three phase unbalanced system
̅𝑰𝒃𝒓,𝒂𝒃𝒄 = 𝐊 𝟑𝚽 ̅𝑰𝒂𝒃𝒄 (2.16)
Step 2: Calculate sum of branch current adjacent to bus 1 of each phase
𝑏𝑟 𝑏𝑟
̅ = ∑𝑖∈𝐴𝑑𝑗𝑎𝑐𝑒𝑛𝑡 𝑡𝑜 𝑏𝑢𝑠 1 𝐼𝑖,𝛹
̅
𝐼1,𝛹 ,∀𝛹 ∈ ℱ (2.17)
Step 3: similar to procedure 1, update 𝐼𝑖𝑏𝑟 but in this case for phase A, B and C
𝑏𝑟
̅ = 𝐼1,𝛹
̅ − (−𝐼1,𝛹
𝑏𝑟 ̅ − 𝐼𝑖,𝛹
𝑏𝑟 ̅ )
𝐼𝑖,𝛹 ,𝑖 = 𝕯(1), ∀𝛹 ∈ ℱ (2.18)
𝑏𝑟
̅ 𝑏𝑟
̅
𝐼(𝑖+1),𝛹 = −𝐼(𝑖+1),𝛹 ,∀𝑖 ∈ 𝕯, ∀𝛹 ∈ ℱ (2.19)
Step 5: Calculate active and reactive branch power generated by (18) and (19)
∗
𝑃𝑖𝑏𝑟,3Ф = ∑𝛹=𝑎,𝑏,𝑐[𝑟𝑒𝑎𝑙(𝑉̅𝑖,𝛹 (𝐼𝑖,𝛹
𝑏𝑟
̅ ) )] ,∀𝑖 ∈ 𝕯 (2.20)
̅ )∗ )]
𝑄𝑖𝑏𝑟,3Ф = ∑𝛹=𝑎,𝑏,𝑐[𝑖𝑚𝑎𝑔(𝑉̅𝑖,𝛹 (𝐼𝑖,𝛹
𝑏𝑟
,∀𝑖 ∈ 𝕯 (2.21)
The 𝑃𝑖𝑏𝑟,3Ф obtained by (2.20) is used to updated 𝑓 3Ф . Respecting to the 𝑓/𝑃 droop characteristic and
the number of the droop bus in IMG, the updated system frequency expression becomes:
𝑓 3Ф = [𝑓𝑖∗ − 𝑚𝑝𝑖 (𝑃𝑖𝑏𝑟,3Ф + 𝑃𝑖𝐿,3Ф )]/𝑛𝐷𝐺 ,∀𝑖 ∈ 𝕯 (2.22)
where 𝑃𝑖𝐿,3Ф denote the active three phase power of the load demand installed at bus 𝑖; and expressed
by:
𝑃𝑖𝐿,3Ф = ∑𝛹=𝑎,𝑏,𝑐 𝑃𝑖,𝛹

𝐿
,∀𝑖 ∈ 𝕯 (2.23)
Then, according to (1) the new updated frequency 𝑓 3Ф is used to update the new active droop three-
phase power generation 𝑃𝑖𝐺,3Ф , as well as the reactances 𝐗 𝐛𝐮𝐬,𝟑Ф and 𝐗 𝐛𝐫,𝟑Ф . Then, the new reactive
droop three-phase power generation 𝑄𝑖𝐺,3Ф is updated using the selected voltage phase B of 𝑽
̅ 𝒂𝒃𝒄 from
the inner loop. Thus, respecting to (2.1) and (2.2) the generation power 𝑃𝑖𝐺,3Ф and 𝑄𝑖𝐺,3Ф expression
becomes:
31
𝑝
𝑃𝑖𝐺,3Ф = (𝑓 ∗ − 𝑓 3Ф )/𝑚𝑖 ,∀𝑖 ∈ 𝕯 (2.24)
𝑞
𝑄𝑖𝐺,3Ф = (𝑉𝑖∗ − |𝑉̅𝑖𝑏 |)/𝑛𝑖 ,∀𝑖 ∈ 𝕯 (2.25)
Comparing to the balanced case, the droop generation power from (2.1) and (2.2) cannot be used
𝐺,(𝑎,𝑏,𝑐) 𝐺,(𝑎,𝑏,𝑐)
directly, due to the requirement of 𝑃𝑖 and 𝑄𝑖 . In fact, there is a need for a methodology
to split the three phases power generation from (2.24) and (2.25) into the respective phases generation.
In this work, the DGs are assumed to have negative sequence compensators that mitigate the voltage
unbalances at the PCC [72] (i.e., the DGs produce balanced voltages despite the system unbalance).
Similar, assumption has been also made in previous three-phase IMG power flow studies (e.g., [15],
[44]). Moreover, the summation of the three-phase power generation should respect the droop
relations. These characteristics can be expressed as following:
|𝑉̅𝑖𝑎 | = |𝑉̅𝑖𝑏 | = |𝑉̅𝑖𝑐 | ,∀𝑖 ∈ 𝕯 (2.26)
2𝜋 2𝜋
𝛿𝑖𝑎 = 𝛿𝑖𝑏 + ( 3 ) = 𝛿𝑖𝑐 − ( 3 ) ,∀𝑖 ∈ 𝕯 (2.27)
𝑃𝑖𝐺,3Ф = 𝑃𝑖𝐺,𝑎 + 𝑃𝑖𝐺,𝑏 + 𝑃𝑖𝐺,𝑐 ,∀𝑖 ∈ 𝕯 (2.28)
𝑄𝑖𝐺,3Ф = 𝑄𝑖𝐺,𝑎 + 𝑄𝑖𝐺,𝑏 + 𝑄𝑖𝐺,𝑐 ,∀𝑖 ∈ 𝕯 (2.29)
Equations (2.26) to (2.29) were previously formulated as mismatch equations in [15] and solved
among the system power flow problem using a newton trust region method. However, there is no
direct way for these equations to be implemented in the branch based unbalanced power flow that we
are adopting in this work. Thus, as can see in Figure 2. 2 the middle loop is used. This loop allows
𝐺,(𝑎,𝑏,𝑐) 𝐺,(𝑎,𝑏,𝑐)
for the determination of 𝑃𝑖 and 𝑃𝑖 respecting to (2.26) to (2.29) and thus capturing the
DG’s behavior in unbalanced three-phase conditions. The expressions for determining each phase
generation power in the middle loop can be given as:
𝑖𝑡 𝑀 −1
𝜕𝑃𝑖𝑏 𝜕𝑃𝑖𝑏
𝑖𝑡 𝑀
0 0 𝑖𝑡 𝑀 −1
𝑖𝑡 𝑀 −1 𝜕𝛿𝑖𝑏 𝜕𝑉𝑖𝑏 2𝜋
𝑃𝑖𝐺,𝑏 𝑃𝑖𝑏 𝛿𝑖𝑎 − 𝛿𝑖𝑏 − ( 3 )
𝜕𝑃𝑖𝑐 𝜕𝑃𝑖𝑐
𝑃𝑖𝐺,𝑐 𝑃𝑖𝑐 0 0 2𝜋
= − 𝜆.
𝜕𝛿𝑖𝑐 𝜕𝑉𝑖𝑐 𝛿𝑖𝑎 − 𝛿𝑖𝑐 + ( 3 )
𝑄𝑖𝐺,𝑏 𝑄𝑖𝑏 𝜕𝑄𝑖𝑏
0
𝜕𝑄𝑖𝑏
0 |𝑉̅𝑖𝑎 | − |𝑉̅𝑖𝑏 |
𝜕𝛿𝑖𝑏 𝜕𝑉𝑖𝑏
𝐺,𝑐
[ 𝑄𝑖 ] [ 𝑄𝑖𝑐 ]
𝜕𝑄𝑖𝑐 𝜕𝑄𝑖𝑐 [ |𝑉̅𝑖𝑎 | − |𝑉̅𝑖𝑐 | ]
0 0
[ 𝜕𝛿𝑖𝑐 𝜕𝑉𝑖𝑐 ]
,∀𝑖 ∈ 𝕯 (2.30)
32
𝑖𝑡𝑀 𝑖𝑡𝑀 −1 𝑖𝑡𝑀
𝑃𝑖𝐺,𝑎 𝑃𝑖𝐺,3Ф 𝑃𝑖𝐺,𝑏 + 𝑃𝑖𝐺,𝑐
[ ] = [ 𝐺,3Ф ] − [ 𝐺,𝑏 ] ,∀𝑖 ∈ 𝕯 (2.31)
𝑄𝑖𝐺,𝑎 𝑄𝑖 𝑄𝑖 + 𝑄𝑖𝐺,𝑐
Where 𝑖𝑡𝑀 is the iteration counter of the middle loop, 𝜆 denote the step size and selected small enough
to maintain the convergence of the middle loop (set as 0.01 in this work). Here it is worth highlighting
that equations (2.30) and (2.31) were previously proposed in [44] for splitting the power among the
three phases and are adopted here for the same purpose. Details about the formulation of the elements
of the matrix in (2.30) are provided in Appendix A of this work. The generation power in (2.30) and
(2.31) are updated through the iterations of the middle loop’s until the mismatch 𝛥𝑚𝑖𝑑 is less than a
given tolerance, where 𝛥𝑚𝑖𝑑 is formulated as:
𝛥𝑚𝑖𝑑 = max(|∑𝛹=𝑎,𝑏,𝑐 𝑉̅𝑖,𝛹 |) ,∀𝑖 ∈ 𝕯 (2.32)
̅ 𝟑Ф
Afterward, procedure 4 is used to update 𝑽 𝟏 :
̅ 3Ф
Procedure 4: Calculate reference bus complex voltage 𝑽 𝟏 for the proposed unbalanced power flow
Step 1: Update the magnitude of DGs voltage magnitude using the branch reactive power obtained
by procedure 3
‖𝑉𝑖𝑥 ‖𝑎 = 𝑉𝑖∗ − 𝑛𝑞𝑖 [𝑄𝑖𝐿,3Ф + 𝑄𝑖𝑏𝑟,3Ф ] ,∀𝑖 ∈ 𝕯 (2.33)
𝑎,𝑏,𝑐
̅𝑖
Step 2: Update 𝑉 ̅ 𝒂𝒃𝒄 from the inner loop with new ‖𝑉𝑖𝑥 ‖𝑎,𝑏,𝑐 obtained by (33) and 𝛿𝑖𝑎 from
of 𝑽
the inner loop
̅𝑖 𝑎 ‖𝑉𝑖𝑥 ‖𝑎 𝑐𝑜𝑠 (𝛿𝑖𝑎 ) + 𝑗‖𝑉𝑖𝑥 ‖𝑎 𝑠𝑖𝑛 (𝛿𝑖𝑎 )

𝑉
𝑥 𝑎 𝑎 2𝜋 𝑥 𝑎 𝑎 2𝜋
̅𝑖 𝑏 ] = ‖𝑉𝑖 ‖ 𝑐𝑜𝑠 (𝛿𝑖 − ( 3 )) + 𝑗‖𝑉𝑖 ‖ 𝑠𝑖𝑛 (𝛿𝑖 − ( 3 ))
[𝑉 ,∀𝑖 ∈ 𝕯 (2.34)
̅𝑖 𝑐 𝑥 𝑎 𝑎 2𝜋 𝑥 𝑎 𝑎 2𝜋
𝑉 [‖𝑉𝑖 ‖ 𝑐𝑜𝑠 (𝛿𝑖 + ( 3 )) + 𝑗‖𝑉𝑖 ‖ 𝑠𝑖𝑛 (𝛿𝑖 + ( 3 ))]
𝑎,𝑏,𝑐 𝐺,(𝑎,𝑏,𝑐) 𝐺,(𝑎,𝑏,𝑐)

Step 3: Update the element of 𝑰̅𝒂𝒃𝒄 with new 𝑉
̅𝑖 as well as 𝑃𝑖 and 𝑄𝑖 obtained by the
middle loop; And calculated ̅̅̅̅
𝒅𝑽𝒙,𝒂𝒃𝒄 expressed by:
̅̅̅̅
𝒅𝑽𝒙,𝒂𝒃𝒄 = 𝐙̅𝐛𝐮𝐬,𝟑Ф ̅𝑰𝒂𝒃𝒄 (2.35)
33
Step 4: 𝑉̅1𝑎,𝑏,𝑐 and 𝛿1𝑎,𝑏,𝑐 can be calculated based on the number of droop bus of the vector 𝑽
̅ 𝒂𝒃𝒄 and
̅̅̅̅
𝒅𝑽𝒙,𝒂𝒃𝒄.
𝑉̅1,𝛹 = [∑(𝑉̅𝑖,𝛹 − ̅̅̅̅

𝑑𝑉𝑖,𝛹𝑥
)]/𝑛𝐷𝐺 ,∀𝑖 ∈ 𝕯, ∀𝛹 ∈ ℱ (2.36)
𝛿1,𝛹 = 𝑎𝑛𝑔𝑙𝑒(𝑉̅1,𝛹 ) − 𝑎𝑛𝑔𝑙𝑒(𝑉̅1𝑎 ) ,∀𝛹 ∈ ℱ (2.37)
̅ 𝒂𝒃𝒄
Step 5: store (36) and (37) in the vector 𝑽 𝟏
̅ 3Ф
and convert to 𝑽 𝟏 .
̅ 𝒂𝒃𝒄
𝑽 𝟏 = [|𝑉̅1𝑎 |∠𝛿1𝑎 ; |𝑉̅1𝑏 |∠𝛿1𝑏 ; |𝑉̅1𝑐 |∠𝛿1𝑐 ] (2.38)
̅ 3Ф
𝑽 ̅ 𝒂𝒃𝒄 ̅ 𝒂𝒃𝒄 ̅ 𝒂𝒃𝒄 𝑇
𝟏 = [𝑽𝟏 𝑽𝟏 … 𝑽𝟏 ](1×3.𝑛 (2.39)
𝑏𝑟 )
To note here that the use of the variables and matrix with the superscript “𝑥” is limited to the scope
̅ and ̅̅̅̅
of procedure 4 only and does not affect processes outside procedure 4 (i.e., does not affect 𝑽 𝒅𝑽
of the inner loop and middle loops). Similar to the proposed balanced algorithm, the algorithm ends
when the mismatch 𝛥𝑉1 goes below a prespecified tolerance. As mentioned earlier, procedure 4 is
̅ 3Ф
also used to initialize 𝑽 𝟏 at the pre-initialize stage. However for step 1 and step 2 in procedure 4,
𝐺,(𝑎,𝑏,𝑐) 𝐺,(𝑎,𝑏,𝑐) 𝐺,(𝑎,𝑏,𝑐) 𝐺,(𝑎,𝑏,𝑐)
𝑃𝑖 , 𝑄𝑖 and 𝑄𝑖𝑏𝑟,3Ф need to be specified. Thus, 𝑃𝑖 and 𝑄𝑖 are initialized by
distributing the total demand (neglecting line losses) in the proportion to the droop gains. Equations
(2.40) and (2.41) express the initialization power of each phase power generation:
−1
1
[∑ 𝑝 ]
𝐺,𝑖𝑛𝑖𝑡 𝑚 𝑛 𝐿,𝑠𝑝𝑒𝑐
𝑖 𝑏𝑢𝑠
𝑃𝑖,𝛹 =(
𝑚𝑖
𝑝 ) ∑𝑗=1 𝑃𝑗,𝛹 ,∀𝑖 ∈ 𝕯, ∀𝛹 ∈ ℱ (2.40)
−1
1
[∑ 𝑞 ]
𝐺,𝑖𝑛𝑖𝑡 𝑛 𝑛 𝐿,𝑠𝑝𝑒𝑐
𝑖 𝑏𝑢𝑠
𝑄𝑖,𝛹 =(
𝑛𝑖
𝑞 ) ∑𝑗=1 𝑄𝑗,𝛹 ,∀𝑖 ∈ 𝕯, ∀𝛹 ∈ ℱ (2.41)
𝐿,𝑖𝑛𝑖𝑡
As can be seen in Figure 2. 2, the power term of (2.33) are replaced by 𝑄𝑖,𝛹 and (2.41) in the pre-
initialize stage.
34
2.4 Algorithm Validation and Numerical Results
In this section, balanced and unbalanced radial distribution test systems have been chosen to evaluate
the effectiveness of the proposed algorithm compared to other algorithms available in the literature.
All case studies were performed in MATLAB environment on a PC equipped with 3.40GHz Intel
Core i7 processor with 8GB of RAM.
2.4.1 Balanced Microgrids
For the validation of the proposed algorithm in balanced conditions, the balanced 33-bus and the
10476-bus distribution test systems were used. Test systems description, feeder parameters and
nominal loads are available online at [73]. The DGs locations, droop coefficients and nominal no-
load settings for the 33-bus and the 10476-bus distribution test systems are summarized in
Table G. 1 of Appendix G.
For the balanced case, the validation methodology is divided into two steps. First, the well-known
33-bus system used in [15] and [30] is used to validate the accuracy of the proposed algorithm, by
comparing its results with those obtained using the NTR algorithm of [15] and the modified
backward/forward sweep (MBF) algorithm of [30]. Figure 2. 3 gives the voltage profile obtained
using the three algorithms.
FIGURE 2. 3: Voltage profile for 33-bus IMG system as calculated by NTR [15], MBF [30] and
the proposed algorithm (𝒇= 0.9198 p.u.)
As can be seen in Figure 2. 3, the results from the three algorithms are almost identical. The maximum
difference in 𝑉/𝛿 between the results of the proposed algorithm and the results of NTR algorithm of
[15] was found to be 7.34 × 10−7 /2.14 × 10−5 . Similarly, comparing the results of the proposed
algorithm with those obtained by the MBF algorithm of [30], the maximum difference in 𝑉/𝛿 was
found to be 3.92 × 10−5 /5.61 × 10−4.
35
In terms of computational performance, the number of inner/outer loops iteration required by the
proposed algorithm is 70/23 compared to 327/93 when using the MBF algorithm of [30]. Figure 2. 4
shows the inner/outer loop errors along the required iterations for the proposed algorithm and for the
MBF algorithm of [30]. Figure 2. 5 shows a close up of the iterations error data of the proposed
algorithm presented in Figure 2. 1.
FIGURE 2. 4: Comparison between the convergence characteristics of proposed algorithm and

the MBF algorithm of [30] in terms of inner and outer loops error for the 33-bus IMG system
FIGURE 2. 5: Evolution of inner and outer loops error of the proposed algorithm for the 33-
bus IMG system
On the other hand, the NTR algorithm of [15] converges in only 4 iterations. Despite that, the NTR
algorithm of [15] is computationally more expensive than both the MBF algorithm of [30] and the
proposed algorithm, with an average runtime (computed over 10 runs) of 367.3 ms compared to 69.3
ms and 47.9 ms for the MBF algorithm of [30] and the proposed algorithm, respectively.
36
In the second step of the algorithm validation under balanced conditions, the proposed algorithm is
implemented to solve the PF problem for a large-scale IMG comprising 10476-bus [71]. As
anticipated, the NTR algorithm of [15] and the MBF algorithm of [30] were not able to solve the PF
problem for this large-scale IMG and they both did not converge. For the proposed algorithm, the
number of required iterations was 47/19 of inner/outer loops and 14753.6 ms of average runtime
(computed over 10 runs). The bus voltage of the 10476-bus system is illustrated in Figure 2. 6.
FIGURE 2. 6: Voltage profile of 10476-bus IMG system (𝒇= 0.9952 p.u.)
2.4.2 Unbalanced Microgrids
To show the effectiveness of the proposed algorithm under unbalanced conditions, the unbalanced
25-bus test system is used. Details about the test system feeder parameters and nominal loads can be
found in [74]. Similar to [15], in this work the system loads have been modified to make the system
heavily unbalanced (by decreasing phase A by 20% and increasing Phase C load by 20%). The DGs
locations, droop coefficients and nominal no-load settings for the DG units the unbalanced 25-bus
test system is found in Table G. 1 of Appendix G. The tolerances were set at 10−6 and 10−4 for
outer/inner loop and middle loop, respectively.
First, we validate the accuracy of the proposed algorithm by comparing its results with those obtained
using the NTR algorithm of [15]. Here it is worth noting that the modified backward/forward sweep
(MBF) algorithm of [30] does not consider unbalanced cases and as such was not used in this
comparison. Figure 2. 7 shows the bus voltage profile obtained by the proposed algorithm and that
obtained by using the NTR algorithm of [15].
37
FIGURE 2. 7: Voltage profile for 25-bus unbalanced IMG as calculated by NTR [15] and by the
proposed algorithm (𝒇= 0.9978 p.u.): (a) Phase A, (b) Phase B and (c) Phase C
As can be seen in Figure 2. 7 the results obtained from the proposed algorithm are almost identical to
those obtained using the NTR algorithm of [15]. The maximum difference in 𝑉/𝛿 between the results
of the proposed algorithm and the results of NTR algorithm of [15] was found to be
1.21 × 10−5 /4.22 × 10−3.
In terms of computational performance, the number of inner/middle/outer loops iteration required by

the proposed algorithm is 114/571/49. However, the number of iteration is 5 when using the NTR
algorithm of [15]. Despite that, the NTR algorithm of [15] is computationally more expensive than
the proposed algorithm, with an average runtime (computed over 10 runs) of 3590.08 ms compared
to 686.09 ms for the proposed algorithm, i.e., the proposed algorithm is around six time faster than
the NTR [15] for the 25-bus unbalanced IMG. Figure 2. 8 shows the evolution of the error against
time of case 1 for the 25-bus unbalanced IMG. As can be seen, the middle loop takes the majority of
computation time compared to the inner and outer loops.
38
FIGURE 2. 8: Evolution of inner, middle and outer loops error of the proposed algorithm for
the base case 25-bus system
2.4.3 Microgrids with voltage and frequency dependent loads
In order to illustrate the effectiveness of proposed algorithm in capturing the load dependency on
voltage and frequency, voltage and frequency dependent load models were adopted in the proposed
power algorithm as expressed by [15]:
𝐿 𝐿,𝑠𝑝𝑒𝑐 𝛼
𝑃𝑖,𝛹 = 𝑃𝑖,𝛹 |𝑉̅𝑖,𝛹 | (1 + 𝐾 𝑝𝑓 (∆𝑓) ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝛹 ∈ ℱ (2.42)
𝐿 𝐿,𝑠𝑝𝑒𝑐 𝛽
𝑄𝑖,𝛹 = 𝑄𝑖,𝛹 |𝑉̅𝑖,𝛹 | (1 + 𝐾 𝑞𝑓 (∆𝑓) ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝛹 ∈ ℱ (2.43)
𝐿,𝑠𝑝𝑒𝑐 𝐿,𝑠𝑝𝑒𝑐
where 𝑃𝑖,𝛹 and 𝑄𝑖,𝛹 are the specified nominal active and reactive power demand; ∆𝑓 denote
the frequency deviation of the IMG from the nominal frequency; α and β are the active and reactive
power exponents which are used to represent constant power, constant current, constant impedance
loads [15]; and 𝐾 𝑝𝑓 and 𝐾 𝑞𝑓 are the active and reactive power frequency dependencies factors. Three
cases studies were considered; case 1 represents the base case where all the loads are considered as
constant power loads and independent to changes in voltage and frequency (i.e., similar to the case
studies presented in subsections 4A and 4B). Cases 2A and 2B consider the system loads as constant
current and constant impedance loads, respectively; and both are dependent to the change of
frequency with 𝐾 𝑝𝑓 and 𝐾 𝑞𝑓 set as 1 and -1 respectively. Table 2. 1 summarizes the accuracy
validation, by comparing the maximum error in voltage magnitude, phase angle, active and reactive
power with respect to the results obtained by MBF and NTR, respectively. These results demonstrate
the effectiveness of the proposed balanced and unbalanced system under various load models.
39
TABLE 2. 1: Proposed algorithm maximum errors compared to MBF [30] and NTR [15] for
the 33-bus balanced and 25-bus unbalanced IMG under different load characteristics
Test system Case max|V| max|δ| max|P| max|Q|
(p.u.) (deg.) (p.u.) (p.u.)
MBF and Proposed algorithm
-05
1 3.92 x10 5.61 x10-04 1.10 x10-07 1.34 x10-04
33-bus -06
2A 1.65 x10 4.63 x10-06 1.26 x10-05 1.49 x10-05
system
2B 1.88 x10-05 1.88 x10-04 1.89 x10-04 3.00 x10-04
NTR and Proposed algorithm
1 7.34 x10-07 2.14 x10-05 1.72 x10-04 1.40 x10-04
33-bus
2A 9.54 x10-07 2.05 x10-05 1.09 x10-06 1.02 x10-05
system
2B 2.76 x10-06 2.43 x10-05 5.81 x10-06 3.73 x10-05
1 1.21 x10-05 4.22 x10-03 1.75 x10-04 2.03 x10-04
25-bus
2A 3.73 x10-05 4.78 x10-03 2.96 x10-04 6.32 x10-05
system
2B 5.66 x10-05 6.37 x10-03 4.19 x10-04 6.83 x10-05
In terms of computational performance, Table 2. 2 and Table 2. 3, respectively, summarize the

average runtime (computed over 10 runs) and the number of iterations required for all methods (i.e.
MBF [30], NTR [15] and the proposed algorithm) for cases 1, 2A and 2B.
TABLE 2. 2: Average runtime (s) of proposed algorithm, MBF [30] and NTR [15] under
different load characteristics
Test system Case MBF NTR Proposed algorithm
(ms) (ms) (ms)
1 69.3 367.3 47.9
33-bus
2A 72.8 379.8 48.1
system
2B 75.7 393.3 48.8
1 14753.6
10476-bus
2A - - 14969.3
system
2B 15145.4
1 3590.1 686.1
25-bus
2A - 3578.3 755.4
system
2B 3553.1 797.2
Similar to the results of case 1 (discussed in subsection 2.4.1), the MBF and NTR could not handle
the 10476-bus system in cases 2A and 2B, and both algorithms did not converge for this system. For
the 33-bus balanced and the 25-bus unbalanced IMG, even though the NTR only takes 4 and 5
iterations, respectively, it is found to be around 8 and 6 six time slower than the proposed algorithm.
For the proposed algorithm, in cases 2A and 2B the load frequency and voltage dependencies require
active and reactive power demand calculation to compute the reference voltage in the outer loop as
40
well as the current injection in the inner loop and as such the proposed algorithm require slightly more
time compared to case 1 wherein the load is constant.
TABLE 2. 3: Number of iterations of proposed algorithm, MBF [30] and NTR [15] under
different load characteristics
Test system Case MBF NTR Proposed algorithm
inner/outer iteration inner/outer
1 327/93 4 70/23
33-bus
2A 218/73 4 71/23
system
2B 306/98 4 76/24
1 47/19
10476-bus
2A - - 48/20
system
2B 57/24
iteration inner/middle/outer
1 5 114/571/49
25-bus
2A - 5 117/617/49
system
2B 5 121/673/49
2.5 Conclusion
This chapter proposed a Zbus PF algorithm for IMGs. The proposed algorithm is inversion-free and
does not require bus renumbering. This algorithm overcomes the problems that existing IMG PF
algorithms face when handling a practical large-scale IMG system comprising thousands of buses.
The accuracy and effectiveness of the proposed algorithm has been validated through several case
studies on a 33-bus and 10476-bus balanced test systems. The algorithm is also extended for the
unbalanced IMG cases. The unbalanced 25 bus system was used and solved considering three cases
for the load model i.e. constant power, constant current and constant impedance. The proposed
algorithm shows superior in terms of simulation time and applicability to practical large-scale IMG
cases.
41
Chapter 3 Linear Power Flow Model for MG system
Y. Vilaisarn, M. Abdelaziz, and J. Cros, “Linear Three-Phase Power Flow Model for Autonomous
Microgrid Systems”. Published in CIRED – Open Access Proceedings Journal, April 2021.
Résumé
Dans ce travail, nous proposons un modèle linéaire triphasé pour l’étude d’écoulement de puissance
des micro-réseaux autonomes fonctionnement en mode « statisme ». Ce modèle linéaire adopte un
ensemble d'équations algébriques linéaires et peut, avec de petites erreurs par rapport au modèle non
linéaire, tenir compte de l'absence d'un bus infini, de la variabilité de la fréquence des micro-réseaux
autonomes et du fonctionnement des unités DG en mode « statisme ». Le modèle linéaire proposé
peut donc fournir une solution rapide et précise. Ces qualités sont très utile pour l’étude d’un problème
de planification qui demande une analyse stochastique et aussi les applications temps réel liée aux
fonctionnements des micro-réseaux autonomes. En plus, ce modèle linéaire est très avantageux pour
faire l'analyse linéaire ou convexe des problèmes d’opération optimale et des problèmes probabilistes.
La précision et l'efficacité du modèle linéaire proposé sont validées par différentes études de cas
portant sur des systèmes de test équilibrés et déséquilibrés.
Abstract
In this work, we propose a linear three-phase power flow model for droop-controlled autonomous
microgrids. The proposed linear power flow model adopts a set of linear algebraic equations and can,
with small errors compared to the detailed nonlinear model, account for the lack of a slack bus, the
variability of autonomous microgrid frequency and the operation of DG units in droop control mode.
The proposed linear model can as such provide a fast and accurate power flow solution and has great
potential in the stochastic planning and online operational applications of autonomous microgrids, as
well as in linear or convex optimal power flow and probabilistic load flow analysis. The accuracy and
effectiveness of the proposed linear power flow model is validated though different case studies
considering balanced and unbalanced test systems.
3.1 Introduction
The planning and operation of microgrids require an accurate power flow analysis of the network in
both grid-connected and autonomous/islanded modes of operation. On the other hand, conventional
power flow model are not suitable for the analysis of autonomous microgrids (AMGs) for three main
42
reasons: (i) the absence of a slack bus in AMGs, (ii) the variability of the system frequency in the
AMGs operation mode, where the frequency is one of the power flow variables to be calculated, and
(iii) the operation of the DG units forming the AMG in droop mode, which cannot be modelled in
conventional distribution power flow models [75], [76].
In this chapter, a linear power flow model for autonomous microgrid or IMG is proposed. Linear
power flow (LPF) models can provide fast power flow solutions and enable a wide range of
operational control and stochastic planning algorithms that would otherwise be computationally
infeasible [45]. Moreover, the nature of LPF model is a node-based which can be used in system
optimization problems as a power equation (i.e., planning and optimal operation). The non-linear
power flow (non-LPF) equations for unbalanced droop-controlled AMG are first presented. Then, the
linearization operator is adopted in the order to turn the non-LPF model to an accurate LPF model.
To this end, the accuracy and the effectiveness of the proposed LPF model are validated thought the
solution delivered by the existing method i.e., Newton Trust Region (NTR) [15], Modified Backward
Forward (MBF) [30] and the proposed sparse 𝐙𝐛𝐮𝐬 in chapter 2.
The remainder of this chapter is organized as follows: Section 3.2 introduces the formulation of the
linear model and the methodology for solving them. The validation of the proposed algorithm is
discussed in Section 3.3, where balanced and unbalanced case studies were deployed to illustrate the
effectiveness of the proposed LPF model. Finally, in Section 3.4 we summarize and conclude the
chapter.
3.2 Formulation and Linear Power Flow

In droop-controlled AMGs, the frequency and voltage at a DG unit’s point of common coupling
(PCC) are controlled to follow the droop characteristic equations, 𝑓/𝑃 and 𝑉/𝑄, given as follows:
𝑝
𝑃𝑖𝐺,3Ф = (𝑓𝑖∗ − 𝑓 3Ф )/𝑚𝑖 ,∀𝑖 ∈ 𝕯 (3.1)
𝑞
𝑄𝑖𝐺,3Ф = (𝑉𝑖∗ − |𝑉̅𝑖𝑎 |)/𝑛𝑖 ,∀𝑖 ∈ 𝕯 (3.2)
where 𝑃𝑖𝐺,3Ф and 𝑄𝑖𝐺,3Ф are the active and reactive three-phase generation power delivered by the DG
unit at the bus 𝑖, respectively, 𝑓 3Ф is the AMG’s frequency, 𝑓𝑖∗ and 𝑉𝑖∗ denote the no-load nominal
frequency and voltage set point of droop bus 𝑖, respectively. The active and reactive static droop gains
𝑝 𝑞
are denoted by 𝑚𝑖 and 𝑛𝑖 . Then, 𝑉̅𝑖 denote the voltage at the PCC of the DG unit connected to bus 𝑖.
43
As can be seen, 𝑄𝑖𝐺,3Ф are respected to the droop by referring to 𝑉̅𝑖𝑎 at the PCC of the DG unit
connected to bus 𝑖 [15].
The non-linear power flow equation in unbalanced droop-controlled AMG can be given as [15], [77]:
𝛹𝛽
𝐺 𝐿 𝑛𝑏𝑢𝑠
𝐺𝑖𝑗 cos(𝛿𝑖,𝛹 − 𝛿𝑗,𝛽 )
𝑃𝑖,𝛹 = (𝑃𝑖,𝛹 − 𝑃𝑖,𝛹 ) = |𝑉̅𝑖,𝛹 | ∑𝑗=1 ∑𝛽∈ℱ|𝑉̅𝑗,𝛽 | ( 𝛹𝛽
)
+𝐵𝑖𝑗 sin(𝛿𝑖,𝛹 − 𝛿𝑗,𝛽 )
,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝛹 ∈ ℱ (3.3)
𝛹𝛽
𝐺 𝐿 𝑛𝑏𝑢𝑠
−𝐵𝑖𝑗 cos(𝛿𝑖,𝛹 − 𝛿𝑗,𝛽 )
𝑄𝑖,𝛹 = (𝑄𝑖,𝛹 − 𝑄𝑖,𝛹 ) = |𝑉̅𝑖,𝛹 | ∑𝑗=1 ∑𝛽∈ℱ|𝑉̅𝑗,𝛽 | ( 𝛹𝛽
)
+𝐺𝑖𝑗 sin(𝛿𝑖,𝛹 − 𝛿𝑗,𝛽 )
,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝛹 ∈ ℱ (3.4)
𝐿 𝐿
where 𝛹 and 𝛽 denote phase indices, 𝑛𝑏𝑢𝑠 is the number of buses in the AMG, 𝑃𝑖,𝛹 and 𝑄𝑖,𝛹 are the
𝛹𝛽 𝛹𝛽
active and reactive load demand of phase 𝛹 located at bus 𝑖, respectively, 𝐺𝑖𝑗 and 𝐵𝑖𝑗 denote the
self/mutual conduction and susceptance connecting bus 𝑖 to 𝑗, respectively. 𝑉̅ and 𝛿 are the complex
voltage magnitude and its phase angle, respectively. Considering constant power loads, the only non-
linearity in (3.3) and (3.4) came from the production of voltage magnitude and trigonometric i.e. sines
and cosines. In order to linearize these equations, the following linear approximation are adopted:
Approximation 1: deal with the non-linearity obtained by the effect of trigonometric sines, cosines
and the deviation between the phase angle i.e. (𝛿𝑖,𝛹 − 𝛿𝑗,𝛽 ) as follows [46]:
1 , if 𝛹 = 𝛽
1 √3
cos(𝛿𝑖,𝛹 − 𝛿𝑗,𝛽 ) ≈ − 2 − 2 (𝛿𝑖,𝛹 − 𝛿𝑗,𝛹 ), if 𝛽 lag 𝑡𝑜 𝛹 (3.5)
1 √3
− + (𝛿 − 𝛿𝑗,𝛹 ), if 𝛽 lead 𝑡𝑜 𝛹
{ 2 2 𝑖,𝛹
(𝛿𝑖,𝛹 − 𝛿𝑗,𝛹 ) , if 𝛹 = 𝛽
1 √3
𝑠𝑖𝑛(𝛿𝑖,𝛹 − 𝛿𝑗,𝛽 ) ≈ − 2 (𝛿𝑖,𝛹 − 𝛿𝑗,𝛹 ) + 2
, if 𝛽 lag 𝑡𝑜 𝛹 (3.6)
1 √3
{ − 2 (𝛿𝑖,𝛹 − 𝛿𝑗,𝛹 ) − 2 , if 𝛽 lead 𝑡𝑜 𝛹
Approximation 2: deal with the second order of the voltage as well as the relation between the phase
angle deviation and the susceptance term as follows [46]:
𝛹𝛽 𝛹𝛽
|𝑉̅𝑖,𝛹 ||𝑉̅𝑗,𝛽 |𝐻𝑖𝑗 cos(𝛿𝑖,𝛹 − 𝛿𝑗,𝛽 ) ≈ |𝑉̅𝑗,𝛽 |𝐻𝑖𝑗 cos(𝛿𝑖,𝛹 − 𝛿𝑗,𝛽 ) (3.7)
44
𝛹𝛽 𝛹𝛽
|𝑉̅𝑖,𝛹 ||𝑉̅𝑗,𝛽 |𝐻𝑖𝑗 sin(𝛿𝑖,𝛹 − 𝛿𝑗,𝛽 ) ≈ 𝐻𝑖𝑗 sin(𝛿𝑖,𝛹 − 𝛿𝑗,𝛽 ) (3.8)
∑𝑛𝑗=1
𝑏𝑢𝑠 𝛹𝛽 𝛹𝛽
𝐻𝑖𝑗 (𝛿𝑖,𝛹 − 𝛿𝑗,𝛹 ) ≈ 𝐻𝑖𝑗 𝛿𝑗,𝛹 (3.9)
Where 𝐻 can be replaced by 𝐺 or 𝐵 depending on (3.3) and (3.4); Cosines in (3.5) and sines (3.6)
were eliminated by approximation 1. In this work, the frequency was assumed close to the nominal
value and its effect on the lines impedances was neglected. By applying approximations 1 and 2 to
eqs. (3.3) and (3.4), the linearized droop three-phase power flow model for droop-controlled AMG
can be given as:
𝑮𝛹,𝑎:𝑐
𝑖𝑗 ∘ 𝑹1𝛹,𝑎:𝑐 . 𝑽𝑗𝑎:𝑐
( )
𝑛 +𝑮𝛹,𝑎:𝑐
𝑖𝑗 . 𝑹𝛹,𝑎:𝑐
2 𝜹𝑗𝑎
𝑃𝑖𝐺,𝛹 − 𝑃𝑖𝐿,𝛹 = ∑𝑗=1
𝑏𝑢𝑠
(3.8)
−𝑩𝛹,𝑎:𝑐
𝑖𝑗 ∘ 𝑹1𝛹,𝑎:𝑐 . 𝜹𝑗𝑎:𝑐
+( )
[ +𝑩𝛹,𝑎:𝑐
2 ]
𝑩𝛹,𝑎:𝑐
−( )
𝑛 +𝑩𝛹,𝑎:𝑐
2 𝜹𝑗𝑎
𝑄𝑖𝐺,𝛹 − 𝑄𝑖𝐿,𝛹 = ∑𝑗=1
𝑏𝑢𝑠
(3.9)
−𝑮𝛹,𝑎:𝑐
+( )
[ +𝑮𝛹,𝑎:𝑐
2 ]
𝑽𝑗𝑎:𝑐 = [𝑉𝑗𝑎 𝑉𝑗𝑏 𝑉𝑗𝑐 ] and 𝜹𝑗𝑎:𝑐 = [𝛿𝑗𝑎 𝛿𝑗𝑏 𝛿𝑗𝑐 ]. In this work, “∘” and “.” represent the Hadamard and
scalar multiplication respectively. 𝑹1𝛹,𝑎:𝑐 and 𝑹𝛹,𝑎:𝑐

2 are the row vector of rotation based on cosine
and sines of the phase deviation. The matrices contained these vectors are:
1 −0.5 −0.5
𝐑𝑎:𝑐,𝑎:𝑐
1 = [−0.5 1 −0.5] (3.10)
−0.5 −0.5 1
0 0.866 −0.866
2 = [−0.866 0 0.866 ] (3.11)
0.866 −0.866 0
Similar to [15], in this work the voltage angle of phase A at bus #1 is taken as the reference and set
by 𝛿1𝑎 = 0, which allows for the consideration of the system frequency as one of the power flow
variables while maintaining the number of equations equal to the number of variables. The DG units
are also assumed to have negative sequence compensators that mitigate the voltage unbalances at the
PCC (i.e., the DGs produce balanced voltages despite the system unbalance) [15]. Moreover, the
45
summation of the three-phase power generation should respect the droop relations. These
characteristics can be expressed as following:
|𝑉̅𝑖𝑎 | = |𝑉̅𝑖𝑏 | = |𝑉̅𝑖𝑐 | (3.12)
2𝜋 2𝜋
𝛿𝑖𝑎 = 𝛿𝑖𝑏 + 3
= 𝛿𝑖𝑐 − 3
(3.13)
𝑃𝑖𝐺,𝑎 = 𝑃𝑖𝐺,3Ф − 𝑃𝑖𝐺,𝑏 − 𝑃𝑖𝐺,𝑐 (3.14)
𝑄𝑖𝐺,𝑎 = 𝑄𝑖𝐺,3Ф − 𝑄𝑖𝐺,𝑏 − 𝑄𝑖𝐺,𝑐 (3.15)
3.3 Validation of the Algorithm

In the order to evaluate the effectiveness of the proposed LPF model, balanced an unbalanced radial
distribution have been used and the results of the developed model were compared with the results of
the nonlinear models in [15] and [30], denoted NTR and MBF, respectively.
Here it is worth noting that the MBF algorithm of [30] does not consider unbalanced cases. All case
studies were performed in MATLAB environment on a PC equipped with 2.50GHz Intel Core i7
processor with 8GB of RAM. In this work, the well-known 33-bus system [15], [31] and the large
scale 10476-bus [73] were used for the simulation under balanced criteria; while in the case of
unbalanced system 25-bus [15] system has been chosen. Similar to [15], in this work the system loads
of 25-bus have been modified to make the system heavily unbalanced (by decreasing phase A by 20%
and increasing Phase C load by 20%). The tolerances were set at 10-6 for NTR and similar value to
MBF.
The DG units’ configuration and location are given in Table G. 1 of Appendix G. First, to illustrate
the computation performance, Table 3. 1 summarize the average runtime (computed over 10 runs)
and the number of iterations of required for NTR [15] and MBF [30].
TABLE 3. 1: Average runtime (ms) and number of iterations of proposed linear model, MBF
[30] and NTR [15]
Test system MBF NTR Proposed LPF
(ms) (inner/outer) (ms) (iter.) (ms)
33-bus 112.6 327/93 516.2 5 8.2
10497-bus un-convergence un-convergence 102052.2
25-bus unsupportable 3414.1 5 30.9
46
It is worth noting that the proposed LPF is solved by matrix inversion and as such does not require
any iterations to solve the power flow, which explained its superior computational performance
compared to NTR and MBF under both balanced or unbalanced conditions. In the case of the large-
scale AMG, both MBF [30] and NTR [15] could not handle the system dimensions and end up with
the un-convergence of the simulation.
In the order to evaluate the accuracy of the proposed LPF, Table 3. 2 summarize the accuracy
validation, by comparing the error indices i.e. maximum error in voltage magnitude, phase angle,
power generation and total losses with respect to the result derived by MBF [30] and NTR [15].
TABLE 3. 2: Proposed LPF error indices compared to MBF [30] and NTR [15] for 33-bus
balanced and 25-bus unbalanced AMG.
Error indices 33-bus 25-bus
Balanced system Unbalanced system
MBF & LPF NTR & LPF NTR & LPF
|V|max (p.u.) 7.84 x10 -4
8.25 x10 -4
3.48 x10-3
|δ|max (deg.) 2.98 x10 -2
3.02 x10 -2
1.32 x10-1
|PG|max (p.u.) 1.86 x10-2 1.85 x10-2 4.04 x10-3
|QG|max (p.u.) 1.27 x10 -2
1.34 x10 -2
3.48 x10-3
Ploss,total (p.u.) 9.72 x10-4 9.77 x10-4 9.08 x10-4
Qloss,total (p.u.) 1.49 x10 -3
1.48 x10 -3
1.02 x10-4
Then, Figure 3. 1 and Figure 3. 2 illustrate the comparison of voltage profile between proposed LPF
and the other algorithms (i.e. MBF [30] and NTR [15]) under balanced and unbalanced, respectively.
FIGURE 3. 1: Voltage profile for 33-bus AMG system as calculated by MBF [30], NTR [15]
and the proposed LPF (𝒇𝑴𝑩𝑭 ≈ 𝒇𝑵𝑻𝑹 = 0.9198 and 𝒇𝑳𝑷𝑭 = 0.9208)
47
FIGURE 3. 2: Voltage profile for unbalanced 25-bus unbalanced IMG as calculated by NTR
[15] and by the proposed LPF (𝒇𝑵𝑻𝑹,𝟑Ф = 0.9978 and 𝒇𝑳𝑷𝑭,𝟑Ф = 0.9981): (a) Phase A, (b) Phase
B and (c) Phase C
Despite its computational superiority, and as can be expected, there is some deviation between the
results obtained by the proposed LPF model and those obtained by MBF [30] and NTR [15] due to
the linearization approximations applied. Nevertheless, the deviations were limited under the range
of 10−3~10−4.
Figure 3. 3 shows the voltage profile of the 10476-bus droop-controlled AMG. This result
demonstrates the capability of the proposed LPF to handle the large-scale droop-controlled AMG
system comprising up to a thousand bus, where NTR [15] and MBF [30] fail to converge under these
conditions.
FIGURE 3. 3: Voltage profile of 10476-bus AMG system (𝒇= 0.9952)
48
Finally, in the order to validate the results for the case 10476-bus delivered by the proposed LPF, the
proposed Zbus method in chapter 2 is used. In this case, the maximum difference in voltage between
both method is found at 1.42 × 10−4 p.u., while in term of the phase angle the maximum difference
is 1.95 × 10−3 degree.
3.4 Conclusion
A new LPF model for droop-controlled AMG was proposed in this work. The proposed model
respects the droop-controlled AMG system special philosophy of operation and allow for the
consideration of the unbalanced conditions typically present in practical distribution systems. The
accuracy and effectiveness of the proposed algorithm has been validated through several balanced
and unbalanced case studies applied on the 33-bus, 10476-bus and 25-bus test systems and comparing
its results to those of two state-of-the art algorithms in [15] and [30]. The results demonstrated the
computational superiority of proposed LPF model and its ability to provide fast power flow solutions.
The deviation between the proposed LPF model solutions and those obtained by the nonlinear models
of the literature were found to be under the range of 10−3 ~10−4 . The proposed LPF is also able to
handle a large-scale network that existing nonlinear model cannot solve. The proposed linear model
can be effective in enabling linear optimization applications for droop-controlled AMGs.
49
Chapter 4 MILP Model for MG’s Optimum Operation
Y. Vilaisarn, M. Moradzadeh, and M. Abdelaziz, “An MILP formulation for the optimum operation
of AC microgrids with hierarchical control”. Published in International Journal of Electrical Power
Energy Systems, September 2021.
Résumé
Un modèle linéaire qui sert à déterminer le fonctionnement a l’état optimal du micro-réseau est une
clé de solution important pour des nombreuses applications en temps réel et stochastiques.
Néanmoins, nombreux existant modèles ne prennent pas en compte les détails du contrôle
hiérarchique nécessite pour opérer le micro-réseau. Cet omet peut conduire à des solutions infaisables
et sous-optimales. Dans ce travail, un nouveau modèle de programmation linéaire en nombres entiers
(i.e., MILP) pour déterminer le fonctionnement à l’état optimal des micro-réseaux triphasés avec
contrôle hiérarchique est proposé. Tout d'abord, le modèle non linéaire est dérivé pour tenir compte
de la philosophie de fonctionnement du micro-réseau à l’état connecté et l’état isolé du réseau
principal. Ensuite, une approche de linéarisation est adoptée à ce modèle non linéaire afin de le
transformer à un modèle linéaire MILP. L'efficacité du modèle linéaire proposé est validée par
plusieurs études de cas portant sur des systèmes de test équilibrés et déséquilibrés. De plus, une
évaluation détaillée des erreurs est menée pour justifier le modèle proposé.
Abstract
An accurate linear model for the optimal operation of the microgrid is a key enabling solution for
many real time and stochastic applications. Nonetheless, many of the existing microgrid’s optimal
power flow models fall short in considering the details of the microgrid hierarchical control, which
can lead to infeasible and suboptimal solutions. In this work, a new mixed integer linear programing
(MILP) model for the optimal operation of three phase microgrids with hierarchical control is
proposed. First, the nonlinear formulation is derived to account for the microgrid philosophy of
operation in both grid connected and islanding modes of operation. Subsequently, a linearization
approach is adopted to the derived nonlinear model to transform it into the required MILP
formulation. The effectiveness of the proposed model is validated through several case studies
considering balanced and unbalanced test systems. Additionally, a detailed error assessment is
conducted to justify the proposed model.
50
4.1 Introduction
A microgrid can be disconnected from the main grid and operated in the islanding mode due to
intentional (scheduled) or unintentional events (unscheduled). In this islanding mode of operation,
the microgrid’s local generation and energy storage resources are controlled to supply its demand. On
the other hand, a microgrid can also be operated in the grid connected mode to provide ancillary
services to the main grid in its upstream. To optimally control the operation of the microgrid, the
hierarchical control structure is adopted. As explained in the literature in chapter 1, the hierarchical
control structure comprises multiple layers. Typically, the lower-level and intermediate level are
employed for controlling the DG units and for maintaining the microgrid’s frequency and voltage
within the permissible range. On the other hand, higher-level deal with the economic and technical
objectives, wherein the optimization of the operation of the microgrid resources and loads is based
on the required AC optimal power flow (ACOPF) problem which is discussed in this chapter.
In chapter 1, it was shown that there is a need to account for the operation and control of the microgrid
with primary and supervisory secondary control in the ACOPF model in order to deliver accurate and
feasible solutions. In this sense, seeking to overcome the mentioned limitations of the existing
literature, in this chapter we seek to develop an ACOPF formulation for AC microgrids with
hierarchical control to simultaneously consider: 1) both grid-connected and islanded modes of
operation; 2) balanced and unbalanced operation of microgrids; 3) the primary and secondary control
layers of the DG units in the islanded mode; 4) general models of dispatchable resources, non-
dispatchable resources and the energy storage; 5) the time coupling constraints of the resources and
storage located in the microgrid. To this end, first we develop a new MINLP formulation for the
optimal operation of AC microgrids with hierarchical control. Subsequently, the proposed MINLP
formulation is linearized and a MILP formulation is extracted, which can be solved by commercial
optimization solvers.
The remainder of this chapter is organized as follows: First, the proposed MINLP formulation to
model the optimal operation of the unbalanced AC microgrid is presented in Section 4.2.
Subsequently, proposed MINLP model is linearized and the proposed MILP model is introduced in
Section 4.3. The numerical simulation results are given in Section 4.4. Finally, Section 4.5 concludes
the chapter.
51
4.2 MINLP Problem Formulation
In this section, the problem for the optimal operation of a three phase microgrid is formulated as an
MINLP problem. The proposed formulation accounts for both islanded and grid connected modes of
operation, wherein the operation mode is defined by the status of the island isolating switch (ISW)
located at the microgrid’s PCC as shown in Figure 4. 1. 𝑏𝑡𝑚 denotes the status of ISW along the
planning horizon 𝛺T; if 𝑏𝑡𝑚 = 0, the microgrid operates in grid-connected mode, and if 𝑏𝑡𝑚 = 1, the
microgrid operates in islanding mode. This allows the problem to consider the objective function and
constraints pertaining to the microgrid’s operation mode. In this work, the load and the generation are
modelled using the method described in Appendix B [78]. The objective function and the related
constraints for the MINLP model are mathematically formulated in the following subsections.
MICROGRID kW
SUBSTATION PCC
Main Grid
MW (BUS1) PiL , QiL PiL , QiL
MVAR kW
kVAR
TRmain ISW
BUSBAR BUSi
(PCCi)
kW
kVAR
kW
DDG kVAR
FIGURE 4. 1: Basic scheme of the microgrid
4.2.1 Objective function
The objective function given by (4.1) is comprised of four terms aiming to minimize the total
operation cost over the planning horizon 𝛺𝑇 . The first term in (4.1) represents the cost of power
exchange between the microgrid and the main grid as determined by (4.2). The second term, given
by (4.3), refers to the fuel cost consumed by the DDG units [79]. These two terms are restricted for
the grid connected mode of the microgrid. The third term represents the cost of load curtailment
during islanding mode of the microgrid, given by (4.4) [26]. The last term characterizes the voltage
violation index used for improving the voltage quality of the microgrid during both operation modes
by (4.5).
min ∑𝑡∈𝛺𝑇 [(𝐶𝑡𝑃𝐶𝐶 + 𝐶𝑡𝐷𝐷𝐺 )(1 − 𝑏𝑡𝑚 ) + 𝐶𝑡𝐶𝑈𝑅𝑇 𝑏𝑡𝑚 + 𝐶𝑡𝑉𝐼 ] (4.1)
𝑿𝑀𝐼𝑁𝐿𝑃,3𝛷
𝑃𝐶𝐶,3𝛷
𝐶𝑡𝑃𝐶𝐶 = 𝑆𝑏𝑎𝑠𝑒 (𝜎𝑡𝑃𝐶𝐶 ∑𝑖∈Ω𝑃𝐶𝐶 𝑃𝑖,𝑡 )∆𝑡 ,∀𝑡 ∈ 𝛺𝑇 (4.2)
52
𝐷𝐷𝐷,3𝛷 𝑓𝑢𝑒𝑙
𝐶𝑡𝐷𝐷𝐺 = 𝑆𝑏𝑎𝑠𝑒 (∑𝑖∈Ω𝐷𝐷𝐺 𝐶𝑅𝑖𝐷𝐷𝐺 𝑃𝑖,𝑡 𝜎 )∆𝑡 ,∀𝑡 ∈ 𝛺𝑇 (4.3)
𝐶𝑡𝐶𝑈𝑅𝑇 = 𝑆𝑏𝑎𝑠𝑒 (𝜎 𝑐𝑢𝑟𝑡 ∑𝑖∈Ω𝐵𝑈𝑆 𝑏𝑐𝑢𝑟𝑡 𝑙𝑜𝑎𝑑 𝐿,3𝛷

𝑖,𝑡 𝑤𝑖 𝑃𝑖,𝑡 )∆𝑡 ,∀𝑡 ∈ 𝛺𝑇 (4.4)
(1 − 𝑏𝑡𝑚 )𝑉𝑠𝑝𝑒𝑐
𝐺𝐶
+ 𝑏𝑡𝑚 𝑉 ∗∗
𝐶𝑡𝑉𝐼 = 𝑉𝑏𝑎𝑠𝑒 (𝑃𝐶 𝑉𝐼 [∑𝑖∈𝛺𝐵𝑈𝑆 ∑𝛹∈ℱ (| |)]) ∆𝑡 ,∀𝑡 ∈ 𝛺𝑇 (4.5)
−𝑉𝑖,𝛹,𝑡
4.2.2 Constraints
The inequality (4.6) is adopted to have the voltage magnitude of different buses restricted to voltage
magnitude constraints. Eq. (4.7) ensures the quality of power supply by limiting the load curtailment
duration. Then, inequality (4.8) is used to limit the number of load shedding switching cycles for each
𝑐𝑢𝑟𝑡
load at bus 𝑖. Finally, the inequality (4.9) ensures 𝑏𝑖,𝑡 activate only for islanded mode.
𝑉 𝑚𝑖𝑛 ≤ 𝑉𝑖,𝛹,𝑡 ≤ 𝑉 𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.6)
∑𝑡∈𝛺𝑇 𝑏𝑐𝑢𝑟𝑡
𝑖,𝑡 ≤ 𝑇𝑐𝑢𝑟𝑡
𝑖 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 (4.7)
∑𝑡∈𝛺𝑇 |𝑏𝑐𝑢𝑟𝑡 𝑐𝑢𝑟𝑡 𝑐𝑢𝑟𝑡

𝑖,𝑡 − 𝑏𝑖,𝑡−1 | ≤ 2𝑁𝑖 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 (4.8)
𝑐𝑢𝑟𝑡
0 ≤ 𝑏𝑖,𝑡 ≤ 𝑏𝑡𝑚 ,∀𝑖, 𝑘 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ 𝛺𝑇 (4.9)
In the following subsections, the constraints corresponding to the modeling of the microgrid, and the
components located inside and outside the microgrid (i.e., DDG, ESS, and substation) are described.
4.2.2.1 Power flow constraints
The three-phase power flow model is used to model the microgrid given by (4.10) to (4.13). The
model takes into account the mutual inductance and interphase capacitance among the different phase
[46].
𝑃𝐶𝐶 𝐷𝐷𝐺
𝑃𝑖,𝛹,𝑡 + 𝑃𝑖,𝛹,𝑡
𝛹𝛽 𝛿𝑖,𝛹,𝑡 𝑊𝑇,3𝛷
𝑃𝑖,𝑡 𝑃𝑉,3𝛷
+ 𝑃𝑖,𝑡
𝐺𝑖𝑗 cos ( )
−𝛿𝑖,𝛽,𝑡 𝑑𝑐ℎ,3𝛷
𝑃𝑖,𝛹,𝑡 = 𝑉𝑖,𝛹,𝑡 ∑𝑗∈Ω𝐵𝑈𝑆 ∑𝛽∈ℱ 𝑉𝑖,𝛹,𝑡 = +( +𝑃𝑖,𝑡 )⁄3
𝛹𝛽 𝛿𝑖,𝛹,𝑡
+𝐵𝑖𝑗 sin ( ) 𝑐ℎ,3𝛷
−𝜂𝐸𝑆𝑆𝑖 𝑃𝑖,𝑡
−𝛿𝑖,𝛽,𝑡
( ) 𝐿
( −𝑃𝑖,𝛹,𝑡 (1 − 𝑏𝑡𝑐𝑢𝑟𝑡 ) )
,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.10)
53
𝛹𝛽 𝛿𝑖,𝛹,𝑡 𝑃𝐶𝐶 𝐷𝐷𝐺
−𝐵𝑖𝑗 cos ( ) +𝑄𝑖,𝛹,𝑡 + 𝑄𝑖,𝛹,𝑡
−𝛿𝑖,𝛽,𝑡 𝑊𝑇,3𝛷 𝐸𝑆𝑆,3𝛷
𝑄𝑖,𝛹,𝑡 = 𝑉𝑖,𝛹,𝑡 ∑𝑗∈Ω𝐵𝑈𝑆 ∑𝛽∈ℱ 𝑉𝑖,𝛹,𝑡 = (+ (𝑄𝑖,𝑡 + 𝑄𝑖,𝑡 )⁄3)
𝛹𝛽 𝛿𝑖,𝛹,𝑡
+𝐺𝑖𝑗 sin ( ) 𝐿 𝑐𝑢𝑟𝑡
−𝑄𝑖,𝛹,𝑡 (1 − 𝑏𝑡 )
−𝛿𝑖,𝛽,𝑡
( )
,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.11)
𝛹𝛽
𝑔𝑖(𝑘)𝑗(𝑘) cos(𝛿𝑖(𝑘),𝛹,𝑡 − 𝛿𝑗(𝑘),𝛹,𝑡 )
𝑏𝑟
𝑃𝑘,𝛹,𝑡 = 𝑉𝑖(𝑘),𝛹,𝑡 ∑𝛽∈ℱ(𝑉𝑖(𝑘),𝛽,𝑡 − 𝑉𝑗(𝑘),𝛽,𝑡 ) ( 𝛹𝛽 )
+𝑏𝑖(𝑘)𝑗(𝑘) sin(𝛿𝑖(𝑘),𝛹,𝑡 − 𝛿𝑗(𝑘),𝛹,𝑡 )
,∀𝑘 ∈ Ω𝐿𝐼𝑁 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.12)
𝛹𝛽
𝑔𝑖(𝑘)𝑗(𝑘) sin(𝛿𝑖(𝑘),𝛹,𝑡 − 𝛿𝑗(𝑘),𝛹,𝑡 )
𝑏𝑟
𝑄𝑘,𝛹,𝑡 = 𝑉𝑖(𝑘),𝛹,𝑡 ∑𝛽∈ℱ(𝑉𝑖(𝑘),𝛽,𝑡 − 𝑉𝑗(𝑘),𝛽,𝑡 ) ( 𝛹𝛽 )
−𝑏𝑖(𝑘)𝑗(𝑘) cos(𝛿𝑖(𝑘),𝛹,𝑡 − 𝛿𝑗(𝑘),𝛹,𝑡 )
,∀𝑘 ∈ Ω𝐿𝐼𝑁 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.13)
2 2
𝑏𝑟
√(𝑃𝑘,𝛹,𝑡 𝑏𝑟
) + (𝑄𝑘,𝛹,𝑡 ) ≤ 𝐼𝑘𝑏𝑟,𝑚𝑎𝑥 𝑉𝑖(𝑘),𝛹,𝑡 ,∀𝑘 ∈ Ω𝐿𝐼𝑁 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.14)
The constraints (4.10) and (4.11) defines the power injected to the 𝑖 𝑡ℎ bus of the microgrid which are
the well-known node-based three phase power flow equations. Then, (4.12) and (4.13) define
respectively the active and reactive power flowing along the branch 𝑘 in the microgrid [15]. It is
worth noting that, 𝑖(𝑘) and 𝑗(𝑘) denote bus 𝑖 located at the upstream of branch 𝑘 and bus 𝑗 for the
downstream of the branch 𝑘, respectively. Eq. (4.14) is the quadratic constraints representing the
limitation over the transmission line ampacity.
4.2.2.2 DDG operation constraints
The DDG units are typically interfaced with power electric converters and passive output filters [15].
In this study, the hierarchical control structure is employed to control the operation of DDG units.
Figure 4. 2(a) illustrates the basic scheme of the DDG unit with hierarchical control. The control
structure involves multi-layers i.e., internal voltage/current control, primary droop control, and
secondary control layers. The synchronous reference frame phase-locked loop (SRF-PLL) is used to
measure the frequency of the microgrid [16]. More detail about the DDG’s control structure and their
functionality can be found in [49], [80].
54
Voltage MICROGRID MICROGRID
DDG fbus
Resource
control bus Frequency PiL , QiL M1
PCC control bus
Lf kW
To next bus kVAR
in MG
GND Cf BUSj
kW
(PCCj) DDG
kVAR upstream downstream
Driver
fbus
Gn Iabc f Vabc Iabc Vm
PjD DG
PWM 3φ DQ SRF-PLL QjD DG M1
SRF-PLL
θ
I dq V dq ωm
DQ φ θ Ifdq |Vj| δMG
Power f/V Droop
θ calculation restoration control action
V/I D DG D DG PCCj
control P Q uv uω
1/s ω'
ω Secondary
Droop
r ef ω Vdq r ef control action
V dq Ifdq Vdq Control Vdq r ef
Internal loop GND Droop ch s

Primary control Secondary control
(a) (b)
FIGURE 4. 2: (a) Basic scheme of the microgrid (b) Steady-state model of DDG unit’s
operation with primary and secondary control under IS mode
In this work, the steady state model is adopted for dealing with the operation constraints of the DDG
units [15], [16]. Figure 4. 2(b) shows the steady state model for the DDG units operated with the
primary droop and secondary controls. This model is sufficient for determining the electrical variables
of the PCC bus where the DDG unit is connected, while respecting to the DDG unit’s internal circuit
and the control structure.
As can be seen in the Figure 4. 2(b), the operation of the DDG unit, their internal circuits and its
control structure can be modeled as a voltage source where the frequency and the voltage are
controlled via the droop control and restored through the secondary control. Following this definition,
the DGG considering the primary droop and secondary control in the steady state can be modeled as
expressed by (4.15) and (4.16):
𝜔 = 𝜔∗ − 𝑚𝑝 𝑃𝐷𝐷𝐺 + 𝑢𝜔 (4.15)
𝑉 = 𝑉 ∗ − 𝑛𝑞 𝑄 𝐷𝐷𝐺 + 𝑢𝑣 (4.16)
Where, the output signals 𝑢𝜔 and 𝑢𝑣 are delivered through a proportional-integral (PI) controller of
the secondary layer and dependent to the measured frequency 𝜔𝑚 and voltage 𝑉 𝑚 as expressed by:
𝑢𝜔 = 𝑘 𝑝𝜔 (𝜔∗ − 𝜔𝑚 ) + 𝑘 𝑖𝜔 ∫(𝜔∗ − 𝜔𝑚 )𝑑𝑡 (4.17)
55
𝑢𝑣 = 𝑘 𝑝𝑣 (𝑉 ∗∗ − 𝑉 𝑚 ) + 𝑘 𝑖𝑣 ∫(𝑉 ∗∗ − 𝑉 𝑚 )𝑑𝑡 (4.18)
As the proposed problem is formulated to model the steady state condition, the proportional term for
(4.17) and (4.18) can be eliminated ( in steady state: 𝜔∗ − 𝜔𝑚 = 0 and 𝑉 ∗∗ − 𝑉 𝑚 = 0) [16].
Nonetheless, the same approach could not be applied to the integrator part of the (4.17) and (4.18).
Let 𝑢𝜔,𝑖𝑛𝑡 denotes the integrator part of (4.17) and 𝑢𝑣,𝑖𝑛𝑡 for (4.18), as such 𝑢𝜔 = 𝑢𝜔,𝑖𝑛𝑡 and 𝑢𝑣 =
𝑢𝑣,𝑖𝑛𝑡 . As explained in [16], 𝑢𝜔,𝑖𝑛𝑡 can be determined using 𝑢𝜔,𝑖𝑛𝑡 = −𝑘 𝑖𝜔 (𝛿 𝑀𝐺 − 𝜋⁄2) under
steady state condition. On the other hand, in the steady state condition the voltage of the voltage-
controlled bus 𝑉 𝑚 is fixed by 𝑉 ∗∗ = 𝑉 𝑚 and there is no need to consider 𝑉 𝑚 to be a variable in the
proposed problem. This allows us to consider 𝑢𝑣,𝑖𝑛𝑡 to be a variable to be determined in the proposed
optimization problem. As such, the primary droop and secondary control are considered with the
DDG model in the proposed model. Let fbus denotes the 𝑖 𝑡ℎ bus where the secondary frequency
control acts. The DDG unit considering three phase microgrid can be modeled by (4.19) to (4.23):
𝑀𝐺 𝑝 𝐷𝐷𝐺,3𝛷
[− 𝑘𝑖 𝜔 (𝛿𝛹,𝑡 − 𝜋⁄2)⁄𝑚𝑖 ]𝑏𝑡𝑚 + (1 − 𝑏𝑡𝑚 )𝑃𝑖,𝑡 /3, 𝛹 = 𝑎
𝐷𝐷𝐺 𝑀𝐺 𝑝 𝐷𝐷𝐺,3𝛷
𝑃𝑖,𝛹,𝑡 = { [− 𝑘𝑖 𝜔 (𝛿𝛹,𝑡 + 𝜋⁄6)⁄𝑚𝑖 ]𝑏𝑡𝑚 + (1 − 𝑏𝑡𝑚 )𝑃𝑖,𝑡 /3, 𝛹 = 𝑏
𝑀𝐺 𝑝 𝐷𝐷𝐺,3𝛷
[− 𝑘𝑖 𝜔 (𝛿𝛹,𝑡 − 7𝜋⁄6)⁄𝑚𝑖 ]𝑏𝑡𝑚 + (1 − 𝑏𝑡𝑚 )𝑃𝑖,𝑡 /3, 𝛹 = 𝑐
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 (4.19)
𝐷𝐷𝐺 𝑣,𝑖𝑛𝑡 𝑞 𝐷𝐷𝐺,3𝛷

𝑄𝑖,𝛹,𝑡 = [(𝑉𝑖∗ − 𝑉𝑖,𝛹,𝑡 + 𝑢𝑖,𝛹,𝑡 )⁄𝑛𝑖 ]𝑏𝑡𝑚 +(1 − 𝑏𝑡𝑚 )𝑄𝑖,𝑡 /3
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝛹 ∈ ℱ,∀𝑡 ∈ 𝛺𝑇 (4.20)
𝐷𝐷𝐺_3𝛷 𝐷𝐷𝐺
𝑃𝑖,𝑡 = ∑𝛹∈ℱ 𝑃𝑖,𝛹,𝑡 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 (4.21)
𝐷𝐷𝐺_3𝛷 𝐷𝐷𝐺
𝑄𝑖,𝑡 = ∑𝛹∈ℱ 𝑄𝑖,𝛹,𝑡 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 (4.22)
𝑉𝑖,𝛹,𝑡 𝑏𝑡𝑚 = 𝑉𝑖∗∗ 𝑏𝑡𝑚 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 , (4.23)
The constraint (4.19) and (4.20) defines the active and reactive power generated by the DDG units.
The first terms of the constraints refer to the secondary control mode of DDGs in the islanding
operation mode, while their second terms represent the PQ mode for the grid connected operation
mode. The binary input parameter 𝑏𝑡𝑚 is used to activate the terms of (4.19) and (4.20) according to
the microgrid operation mode. In this work, the active and reactive droop gains of the DDG units are
determined using the capacity-based method and expressed by (4.24) and (4.25). By using this
definition, the load in the system can be shared in proportion to the capacity of the DDG units
[26],[54].
56
𝑝
𝑚𝑖 = ∆𝜔⁄𝑃𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 (4.24)
𝑞
𝑛𝑖 = ∆𝑉⁄2Q𝐷𝐷𝐺,𝑚𝑎𝑥
𝑖 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 (4.25)
Then, the constraints (4.21) and (4.22) defines the three-phase active and reactive power flowing to
the bus where the DDG is connected. The constraints (4.23) define the voltage magnitude restored by
the secondary control of the DDGs.
DDG
Feasible operation region
P DDG
DDG
Non-feasible operation
region
S DDG,max
FIGURE 4. 3: Feasible operation region of the DDG units in IS and GC mode
With the interfaced power electronic converter, the DDG units is allowed to control the reactive power
by producing to or receiving from the microgrid [80], [81]. Figure 4. 3 shows the feasible operating
region of the DDG units and the following constraints defines this region accurately.
𝑃𝑖𝐷𝐷𝐺,𝑚𝑖𝑛 ≤ 𝑃𝑖,𝑡
𝐷𝐷𝐺,3𝛷
≤ 𝑃𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 (4.26)
𝑄𝑖𝐷𝐷𝐺,𝑚𝑖𝑛 ≤ 𝑄𝑖,𝑡
𝐷𝐷𝐺,3𝛷
≤ 𝑄𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 (4.27)
𝐷𝐷𝐺,3𝛷
−𝑃𝑖,𝑡 𝑡𝑔(cos −1 𝑃𝐹𝑖𝐷𝐷𝐺,4𝑡ℎ ) ≤ 𝑄𝑖,𝑡
𝐷𝐷𝐺,3𝛷 𝐷𝐷𝐺,3𝛷
≤ 𝑃𝑖,𝑡 𝑡𝑔(cos−1 𝑃𝐹𝑖𝐷𝐷𝐺,1𝑠𝑡 )
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 (4.28)
𝐷𝐷𝐺,3𝛷 2 𝐷𝐷𝐺,3𝛷 2
√(𝑃𝑖,𝑡 ) + (𝑄𝑖,𝑡 ) ≤ 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 (4.29)
where (4.26) and (4.27) limit the active and reactive power while the constraint (4.28) guarantees the
operation of the DDG unit respecting to the minimum power factor set. Finally, (4.29) outlines the
boundary of the feasible region for limiting the active and reactive power with respect to the rating
of the DDG. More details regarding the derivation for the operation feasible region can be found in
the Appendix D.
57
4.2.2.3 Energy storage operational constraints
In this study, a general model is employed for the ESSs that is not limited to any particular energy
storage technology. The ESSs are interfaced with power conversion systems (PCSs) to control the
ESSs active and reactive powers in accordance with the connected network requirements [82].
Accordingly, the ESS can be utilized to provide ancillary services such as voltage support, frequency
regulation, grid stabilization, etc. In this work, the ESSs are supposed to operate at four-quadrant
areas. To this end, the ESS active and reactive powers are limited via the quadratic function as
illustrated in Figure 4. 4.
Inverter unit
kW
kVAR
Q ESS ESS
Feasible operation
region
P ch-P dch
S ESS,max
FIGURE 4. 4: Principle of the ESS units
Considering three phase microgrid, the ESS, therefore, can be model using [82]-[83].
𝑑𝑐ℎ,3𝛷
0 ≤ 𝑃𝑖,𝑡 ≤ 𝑃𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 𝑏𝐸𝑆𝑆
𝑖,𝑡 ,∀𝑖 ∈ Ω𝐸𝐸𝑆 , ∀𝑡 ∈ 𝛺𝑇 (4.30)
𝑐ℎ,3𝛷
0 ≤ 𝑃𝑖,𝑡 ≤ 𝑃𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 (1 − 𝑏𝐸𝑆𝑆
𝑖,𝑡 ) ,∀𝑖 ∈ Ω𝐸𝐸𝑆 , ∀𝑡 ∈ 𝛺𝑇 (4.31)
𝑃𝑖𝐸𝑆𝑆,𝑚𝑖𝑛 ≤ 𝑄𝑖,𝑡
𝐸𝑆𝑆,3𝛷
≤ 𝑄𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝑡 ∈ 𝛺𝑇 (4.32)
𝑑𝑐ℎ,3𝛷 𝑐ℎ,3𝛷 2
𝐸𝑆𝑆,3𝛷 2
√(𝑃𝑖,𝑡 − 𝑃𝑖,𝑡 ) + (𝑄𝑖,𝑡 ) ≤ 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝑡 ∈ 𝛺𝑇 (4.33)
𝐸𝑆𝑆
𝑆𝑂𝐶𝑖0 𝐸𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 + (𝜂𝑖𝐸𝑆𝑆 𝑃𝑖,𝑡
𝑐ℎ,3𝛷 𝑑𝑐ℎ,3𝛷
− 𝑃𝑖,𝑡 )∆𝑡 , 𝑡 = 1
𝐸𝑖,𝑡 ={ 𝐸𝑆𝑆 𝑐ℎ,3𝛷 𝑑𝑐ℎ,3𝛷 ,∀𝑖 ∈ Ω𝐸𝐸𝑆 , ∀𝑡 ∈ 𝛺𝑇 (4.34)
𝐸𝑖,𝑡−1 + (𝜂𝑖𝐸𝑆𝑆 𝑃𝑖,𝑡 − 𝑃𝑖,𝑡 )∆𝑡 ,𝑡 ≥ 2
𝐸𝑖𝐸𝑆𝑆,𝑚𝑖𝑛 ≤ 𝐸𝑖,𝑡
𝐸𝑆𝑆
≤ 𝐸𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐸𝐸𝑆 , ∀𝑡 ∈ 𝛺𝑇 (4.35)
𝐸𝑆𝑆
𝐸𝑖,𝑡 ≥ 𝑆𝑂𝐶𝑖0 𝐸𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐸𝐸𝑆 , 𝑡 = 𝑁𝑇 (4.36)
58
The ESS charging and discharging powers are restricted by their rated power using (4.30) and (4.31).
The output apparent power of the energy storage, connected through an inverter, is restricted by the
maximum apparent power supplied/consumed of the inverter, denoted as 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 (4.33). The ESS
𝐸𝑆𝑆
stored energy in each time segment, 𝐸𝑖,𝑡 , is updated based on its charging/discharging energy flows
and the efficiency of the ESS (4.34). The efficiency 𝜂 𝐸𝑆𝑆 the round trip efficiency of the energy
storage systems [84]-[85]. Although the stored energy level of the ESS is not affected by the
consumed/supplied reactive power of the connected inverter, the amount of active power is dependent
on it due to the line current limitations (4.33)-(4.34). The energy stored in the ESS is limited to its
rated capacity and minimum stored energy (4.35). The ESSs are assumed to be initially charged up
to 𝑆𝑂𝐶𝑖0 𝐸𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 and it must remain equal to or greater than the same amount of energy at the end
of the planning horizon (4.36). In this work, the ESS units are assumed to be operated in power-
controlled mode (PQ mode) formulated by (4.30) and (4.36) similar to the works in [26],[54],[82].
Nevertheless, as the ESS units are interfaced with power electronics inverters, the ESSs units are
capable of operating in droop mode and accordingly with the secondary control. A case study is
provided in subsection 4.4.5 to investigate the ESS operation with primary droop and secondary
control layer.
4.2.2.4 Power exchange constraints
The constraints represent the power exchange between the microgrid, and the main grid can be given
by (4.37) to (4.42). The constraints (4.37) and (4.38) defines the active and reactive three phase power
at the PCC of microgrid. Then, the power imported from the main grid were enforced their limit by
the rating of the transformer located in the substation and given by the quadratic equation (4.39). The
constraint (4.40) limits the operation of the transformer above the minimum power factor allowable.
Moreover, (4.41) and (4.42) specified the voltage magnitude and their phase angle at the PCC which
imposed by the main grid during grid connected mode.
𝑃𝐶C,3𝛷 𝑃𝐶𝐶
𝑃𝑖,𝑡 (1 − 𝑏𝑡𝑚 ) = (1 − 𝑏𝑡𝑚 ) ∑𝛹∈ℱ 𝑃𝑖,𝛹,𝑡 ,∀𝑖 ∈ Ω𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 (4.37)
𝑃𝐶𝐶,3𝛷 𝑃𝐶𝐶
𝑄𝑖,𝑡 (1 − 𝑏𝑡𝑚 ) = (1 − 𝑏𝑡𝑚 ) ∑𝛹∈ℱ 𝑄𝑖,𝛹,𝑡 ,∀𝑖 ∈ Ω𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 (4.38)
𝑃𝐶𝐶,3𝛷 2 𝑃𝐶𝐶,3𝛷 2
√(𝑃𝑖,𝑡 ) + (𝑄𝑖,𝑡 ) ≤ 𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 (4.39)
𝑃𝐶𝐶,3𝛷
−𝑃𝑖,𝑡 𝑡𝑔(cos−1 𝑃𝐹𝑖𝑇𝑅,4𝑡ℎ ) ≤ 𝑄𝑖,𝑡
𝑃𝐶𝐶,3𝛷 𝑃𝐶𝐶,3𝛷
≤ 𝑃𝑖,𝑡 𝑡𝑔(cos −1 𝑃𝐹𝑖𝑇𝑅,1𝑠𝑡 )
,∀𝑖 ∈ Ω𝑃𝐶𝐶 , ∀𝑡 ∈ 𝑁𝑇 (4.40)
59
𝑉𝑖,𝛹,𝑡 (1 − 𝑏𝑡𝑚 ) = 𝑉𝑠𝑝𝑒𝑐
𝐺𝐶 (1
− 𝑏𝑡𝑚 ) ,∀𝑖 ∈ Ω𝑃𝐶𝐶 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.41)
0(1 − 𝑏𝑡𝑚 ), 𝛹 = 𝑎
𝑃𝐶𝐶
𝛿𝑖,𝛹,𝑡 (1 − 𝑏𝑡𝑚 ) = {−2𝜋/3(1 − 𝑏𝑡𝑚 ), 𝛹 = 𝑏 ,∀𝑖 ∈ Ω𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 (4.42)
+2𝜋/3(1 − 𝑏𝑡𝑚 ), 𝛹 = 𝑐
4.3 Linearize Formulation

Solving the proposed MINLP model is computationally expensive due to the nonlinear terms found
in (4.1) to (4.37). In this section, the nonlinear terms are linearized, and the proposed MILP model is
derived.
4.3.1 Linearization Approach
4.3.1.1 Linearization of Absolute Value Operator
Eqs. (4.5) and (4.8) include absolute value operators which make the model non-linear. In order to
linearize (4.5), the two following assumptions are made:
1) During the grid connected mode of the microgrid, the DDG units are operated in PQ mode
and no DDG operates in PV mode in the system similar to the work in [52], [54]. Thus, there
𝐺𝐶
is no other type of resources can specify the voltage rather than 𝑉𝑠𝑝𝑒𝑐 imposed by the upstream
network. In this case, the voltage magnitude for all the buses located at downstream of the
𝐺𝐶
bus which interlinked with the upstream power grid (i.e., bus #1) are lower than 𝑉𝑠𝑝𝑒𝑐 .
2) During the islanded mode of the microgrid, the DDG units are operated with secondary
control layer. Then, a voltage restoration reference 𝑉 ∗∗ is identically set for all DDG units
allocated in the test system. In this case, the voltage of the system attempt to follow 𝑉 ∗∗ .
𝐺𝐶
As there is no bus in the microgrid that its voltage magnitude could overcome 𝑉𝑠𝑝𝑒𝑐 or 𝑉 ∗∗ , the
absolute of voltage deviation can be omitted. Hence, the linearized voltage deviation index (4.5) can
be modified as:
𝑠𝑝𝑒𝑐
(1 − 𝑏𝑡𝑚 )𝑉𝐺𝐶 + 𝑏𝑡𝑚 𝑉 ∗∗
𝐶𝑡𝑉𝐼,𝐿𝑖𝑛 = 𝑉𝑏𝑎𝑠𝑒 (𝑃𝐶 𝑉𝐼
[∑𝑖∈𝛺𝐵𝑈𝑆 ∑𝛹∈ℱ ( )]) ∆𝑡 ,∀𝑡 ∈ 𝛺𝑇 (4.43)
−𝑉𝑖,𝛹,𝑡
Nevertheless, if the aforementioned assumptions made to linearize (4.5) are not applicable, the
method introduced in the Appendix E can be applied.
60
The same linearization method described in Appendix E can be adopted to linearize the absolute value
operator employed in (4.8). In this case, a new binary variable z𝑐𝑢𝑟𝑡
𝑖 is introduced for representing
the term |𝑏𝑐𝑢𝑟𝑡 𝑐𝑢𝑟𝑡
𝑖,𝑡 − 𝑏𝑖,𝑡−1 |. Then, the inequality constraints employed for linearizing (4.8) can be
expressed by:
∑𝑡∈𝛺𝑇 z𝑐𝑢𝑟𝑡
𝑖,𝑡 ≤ 2𝑁𝑐𝑢𝑟𝑡
𝑖 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 (4.44)
𝑏𝑐𝑢𝑟𝑡 𝑐𝑢𝑟𝑡 𝑐𝑢𝑟𝑡

𝑖,𝑡 − 𝑏𝑖,𝑡−1 ≤ z𝑖,𝑡 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ 𝛺𝑇 (4.45)
−𝑏𝑐𝑢𝑟𝑡 𝑐𝑢𝑟𝑡 𝑐𝑢𝑟𝑡

𝑖,𝑡 + 𝑏𝑖,𝑡−1 ≤ z𝑖,𝑡 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ 𝛺𝑇 (4.46)
The constraints (4.45) and (4.46) are adopted to force z𝑐𝑢𝑟𝑡

𝑖,𝑡 to be “1” when there is a transition
between load shedding and power supply for the load in the bus 𝑖. Otherwise, z𝑐𝑢𝑟𝑡
𝑖 is forced to be
zero using (4.44) to (4.46).
4.3.1.2 Linearization of power flow and line flow
In (4.10) and (4.11), with the production of voltage magnitude and trigonometric i.e., sines and
cosines the non-linearity has been introduced in these equations. In this case, two linear
approximation is employed for these equations:
Approximation #1: This approximation technique deals with the non-linearity introduced by the
𝛹𝛽
trigonometric sine and cosine functions and the deviation between the phase angle 𝜃𝑖𝑗 as follows
[46]:
1 , if 𝛹 = 𝛽
cos(𝛿𝑖,𝛹,𝑡 − 𝛿𝑗,𝛽,𝑡 ) ≈ {− 1⁄2 − √3⁄2 (𝛿𝑖,𝛹,𝑡 − 𝛿𝑗,𝛹,𝑡 ), if 𝛽 lag 𝑡𝑜 𝛹
− 1⁄2 + √3⁄2 (𝛿𝑖,𝛹,𝑡 − 𝛿𝑗,𝛹,𝑡 ), if 𝛽 lead 𝑡𝑜 𝛹
,∀𝑖, 𝑗 ∈ Ω𝐵𝑈𝑆 , ∀𝛹, 𝛽 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.47)
𝛿𝑖,𝛹,𝑡 − 𝛿𝑗,𝛹,𝑡 , if 𝛹 = 𝛽
sin(𝛿𝑖,𝛹,𝑡 − 𝛿𝑗,𝛽,𝑡 ) ≈ {− 1⁄2 (𝛿𝑖,𝛹,𝑡 − 𝛿𝑗,𝛹,𝑡 ) + √3⁄2 , if 𝛽 lag 𝑡𝑜 𝛹
− 1⁄2 (𝛿𝑖,𝛹,𝑡 − 𝛿𝑗,𝛹,𝑡 ) − √3⁄2 , if 𝛽 lead 𝑡𝑜 𝛹
,∀𝑖, 𝑗 ∈ Ω𝐵𝑈𝑆 , ∀𝛹, 𝛽 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.48)
Approximation 2: This approximation technique deals with the second order of the voltage as well as
the relation between the phase angle deviation and the susceptance term as follows [46]:
61
𝛹𝛽 𝛹𝛽
𝑉𝑖,𝛹,𝑡 𝑉𝑗,𝛽,𝑡 𝐻𝑖𝑗 cos(𝛿𝑖,𝛹,𝑡 − 𝛿𝑗,𝛽,𝑡 ) ≈ 𝑉𝑗,𝛽,𝑡 𝐻𝑖𝑗 cos(𝛿𝑖,𝛹,𝑡 − 𝛿𝑗,𝛽,𝑡 )
,∀𝑖, 𝑗 ∈ Ω𝐵𝑈𝑆 , ∀𝛹, 𝛽 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.49)
𝛹𝛽 𝛹𝛽
𝑉𝑖,𝛹,𝑡 𝑉𝑗,𝛽,𝑡 𝐻𝑖𝑗 sin(𝛿𝑖,𝛹,𝑡 − 𝛿𝑗,𝛽,𝑡 ) ≈ 𝐻𝑖𝑗 sin(𝛿𝑖,𝛹,𝑡 − 𝛿𝑗,𝛽,𝑡 )
,∀𝑖, 𝑗 ∈ Ω𝐵𝑈𝑆 , ∀𝛹, 𝛽 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.50)
∑𝑛𝑗=1
𝑏𝑢𝑠 𝛹𝛽 𝛹𝛽
𝐻𝑖𝑗 (𝛿𝑖,𝛹,𝑡 − 𝛿𝑗,𝛽,𝑡 ) ≈ 𝐻𝑖𝑗 𝛿𝑗𝛹,𝑡 ,∀𝑖, 𝑗 ∈ Ω𝐵𝑈𝑆 , ∀𝛹, 𝛽 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.51)
where, H can be replaced by 𝐺 or 𝐵 depending on (4.10) and (4.11); Cosine and sine functions within
(4.49) and (4.50) have already been eliminated by Approximation #1. In this work, the frequency of
the system is restored to the nominal value due to the DDGs’ secondary control and its effect on the
lines impedances can be neglected. By applying Approximations #1 and #2, the three-phase power
flow model can be linearized as:
𝑡,𝑠,𝑠𝑒
𝑮𝛹,𝑎:𝑐
𝑖𝑗 ∘ 𝑹1𝛹,𝑎:𝑐 . 𝑽𝑗𝑎:𝑐 𝑃𝐶𝐶
𝑃𝑖,𝛹,𝑡 𝐷𝐷𝐺
+ 𝑃𝑖,𝛹,𝑡
( 𝑡,𝑠,𝑠𝑒 ) 𝑊𝑇,3𝛷 𝑃𝑉,3𝛷
+𝑮𝛹,𝑎:𝑐 . 𝑹𝛹,𝑎:𝑐
2 𝜹𝑗𝛹 𝑃𝑖,𝑡 + 𝑃𝑖,𝑡
𝑃𝑖,𝛹,𝑡 = ∑𝑛𝑗=1
𝑏𝑢𝑠 𝑖𝑗
𝑡,𝑠,𝑠𝑒
= + ( 𝑑𝑐ℎ,3𝛷 𝑐ℎ,3𝛷 )⁄3
𝑖𝑗 ∘ 𝑹1𝛹,𝑎:𝑐 . 𝜹𝑗𝑎:𝑐 +𝑃𝑖,𝑡 − 𝜂𝐸𝑆𝑆𝑖 𝑃𝑖,𝑡
+( ) 𝐿
+𝑩𝛹,𝑎:𝑐 . 𝑹𝛹,𝑎:𝑐 ( −𝑃𝑖,𝛹,𝑡 (1 − 𝑏𝑡𝑐𝑢𝑟𝑡 ) )
( 𝑖𝑗 2 )
,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.52)
𝑡,𝑠,𝑠𝑒
𝑩𝛹,𝑎:𝑐
−( ) 𝑃𝐶𝐶 𝐷𝐷𝐺
𝛹,𝑎:𝑐 𝛹 𝑡,𝑠,𝑠𝑒 +𝑄𝑖,𝛹,𝑡 + 𝑄𝑖,𝛹,𝑡
𝑛
𝑏𝑢𝑠
+𝑩𝛹,𝑎:𝑐
𝑖𝑗 . 𝑹2 𝜹 𝑗 𝑊𝑇,3𝛷 𝐸𝑆𝑆,3𝛷
𝑄𝑖,𝛹,𝑡 = ∑𝑗=1 𝑡,𝑠,𝑠𝑒
= (+(𝑄𝑖,𝑡 + 𝑄𝑖,𝑡 )⁄3)
−𝑮𝛹,𝑎:𝑐 ∘ 𝑹1𝛹,𝑎:𝑐 . 𝜹𝑗𝑎:𝑐 𝐿
+(
𝑖𝑗
) −𝑄𝑖,𝛹,𝑡 (1 − 𝑏𝑡𝑐𝑢𝑟𝑡 )
( +𝑮𝛹,𝑎:𝑐
2 )
,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.53)
where, 𝑽𝑗𝑎:𝑐 = [𝑉𝑗𝑎 𝑉𝑗𝑏 𝑉𝑗𝑐 ] and 𝜹𝑗𝑎:𝑐 = [𝛿𝑗𝑎 𝛿𝑗𝑏 𝛿𝑗𝑐 ]. In this work, “∘” and “.” respectively represent the
Hadamard and scalar multiplications. 𝑹1𝛹,𝑎:𝑐 and 𝑹𝛹,𝑎:𝑐

2 are the row vector of rotation based on cosine
and sines of the phase deviation, respectively. These vectors are as follows:
1 −1/2 −1/2
1 =[−1/2 1 −1/2] (4.54)
−1/2 −1/2 1
62
0 √3/2 −√3/2
2 = [−√3/2 0 √3/2 ] (4.55)
√3/2 −√3/2 0
The same approximation techniques are also applied to the nonlinear constraints (4.12)-(4.13) and
they are linearized as follows:
𝑡,𝑠,𝑠𝑒 𝑡,𝑠,𝑠𝑒
𝑮𝛹,𝑎:𝑐
𝑖𝑗 ∘ 𝑹1𝛹,𝑎:𝑐 . 𝑽𝑎:𝑐
𝑖 𝑮𝛹,𝑎:𝑐
𝑏𝑟
𝑃𝑘,𝛹,𝑡 =− +𝑮𝛹,𝑎:𝑐
2 𝜹𝛹
𝑖 + +𝑮𝛹,𝑎:𝑐
2 𝜹𝑗𝛹
−𝑩𝛹,𝑎:𝑐 ∘ 𝑹1𝛹,𝑎:𝑐 . 𝜹𝑎:𝑐
𝑖 −𝑩𝛹,𝑎:𝑐 ∘ 𝑹1𝛹,𝑎:𝑐 . 𝜹𝑗𝑎:𝑐
( 𝑖𝑗 ) ( 𝑖𝑗
)
,∀𝑘 ∈ Ω𝐿𝐼𝑁 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.56)
𝑖𝑗 ∘ 𝑹1𝛹,𝑎:𝑐 . 𝑽𝑎:𝑐
𝑖 −𝑩𝛹,𝑎:𝑐
𝑏𝑟
𝑄𝑘,𝛹,𝑡 =− −𝑩𝛹,𝑎:𝑐
2 𝜹𝛹
𝑖 + −𝑩𝛹,𝑎:𝑐
2 𝜹𝑗𝛹
−𝑮𝛹,𝑎:𝑐
𝑖𝑗 ∘ 𝑹1𝛹,𝑎:𝑐 . 𝜹𝑎:𝑐
𝑖 −𝑮𝛹,𝑎:𝑐
( ) ( )
,∀𝑘 ∈ Ω𝐿𝐼𝑁 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.57)
4.3.1.3 Linearization of quadratic function
The piecewise linearization technique is applied for the nonlinear quadratic constraints (4.14), (4.29),
(4.33) and (4.39). Specifically in this work, multi-segment polygon region piecewise linearization
technique is employed [86], [87], [88]. This technique focuses on forming multiple constraints
representing the perimeters of circle for restricting the quadratic summation. Figure 4. 5 illustrates
the polygon piecewise technique.
{1𝑠𝑡 ,2nd ,3rd ,4th }

Let’s 𝐿𝑛 and denote the nth piecewise constraints of the quadrants {1𝑠𝑡 , 2nd , 3rd , 4th }.
{1𝑠𝑡 ,2nd ,3rd ,4th }
According to Figure 4. 5, each segment constraints 𝐿𝑛 can be formulated by:
𝑠𝑡 𝑠𝑡 𝑠𝑡
𝐿1𝑠𝑡 1 1 1
𝑛 = {𝛼𝑛 𝑋 + 𝛽𝑛 𝑌 − 𝛾 𝑍 ≤ 0 ,𝑛 = 1,2, … , 𝑛𝑝𝑤 (4.58)
𝐿2𝑛𝑑
𝑛 = {−𝛼𝑛1 𝑋 + 𝛽𝑛1 𝑌 − 𝛾 1 𝑍 ≤ 0 ,𝑛 = 1,2, … , 𝑛𝑝𝑤 (4.59)
𝑡ℎ 𝑡ℎ 𝑡ℎ
𝐿3𝑟𝑑 4 4
𝑛 = {−𝛼𝑛 𝑋 − 𝛽𝑛 𝑌 − 𝛾
4
𝑍≤0 ,𝑛 = 1,2, … , 𝑛𝑝𝑤 (4.60)
𝑡ℎ 𝑡ℎ 𝑡ℎ
𝐿4𝑡ℎ 4 4
𝑛 = {𝛼𝑛 𝑋 − 𝛽𝑛 𝑌 − 𝛾
4
𝑍≤0 ,𝑛 = 1,2, … , 𝑛𝑝𝑤 (4.61)
63
Y
Z
...
Dead zone
L22nd L21s t
2nd Quadrant 1s t Quadrant
L12nd φs el 1s t γ L11s t
φs el 2nd
=φs el 1s t
φ01s t
φs el 3r d
=φs el 4th φ04th X
L1 3r d φs el 4th
L1 4th
rd th
3 Quadrant 4 Quadrant
L23r d L24th
Dead zone
... ...
FIGURE 4. 5: Polygon piecewise linearization plan for the quadratic function approximation
where, 𝑛𝑝𝑤 is the number of piecewise constraints for each quadrant. The 𝑛𝑝𝑤 can be set high to
improve the accuracy. Nonetheless, higher 𝑛𝑝𝑤 increases the number of constraints and the
computation burden is accordingly increased. The piecewise constraint coefficients for each quadrant
can be calculated as following:
𝑠𝑡 ,4𝑡ℎ 𝑠𝑡 ,4𝑡ℎ 𝑠𝑡 ,4𝑡ℎ 𝑠𝑡 𝑡ℎ 𝑠𝑡 𝑡ℎ

𝛼𝑛1 = cos (𝜑𝑛1 + (𝑛1 1 ,4
− 1) 𝜑𝑠𝑒𝑙 ⁄𝑛1𝑝𝑤,4 ) ,𝑛 = 1,2, … , 𝑛𝑝𝑤 (4.62)

𝛽𝑛1 = sin (𝜑𝑛1 + (𝑛1 1 ,4
− 1) 𝜑𝑠𝑒𝑙 ⁄𝑛1𝑝𝑤,4 ) ,𝑛 = 1,2, … , 𝑛𝑝𝑤 (4.63)
𝑠𝑡 ,4𝑡ℎ 1 ,4 𝑠𝑡 𝑡ℎ 𝑠𝑡 𝑡ℎ
𝛾1 = cos (𝜑𝑠𝑒𝑙 ⁄2𝑛1𝑝𝑤,4 ) ,𝑛 = 1,2, … , 𝑛𝑝𝑤 (4.64)
𝑠𝑡 ,4𝑡ℎ 𝑠𝑡 𝑡ℎ 𝑠𝑡 𝑡ℎ
𝜑01 1 ,4
= 𝜑𝑠𝑒𝑙 ⁄2𝑛1𝑝𝑤,4 ,𝑛 = 1,2, … , 𝑛𝑝𝑤 (4.65)
By applying the piecewise constraints to the quadratic functions of Section 4.2 and respecting to the
operation characteristic of the equipment, the formation of linear piecewise constraint associated with
each equipment is summarized in Table 4. 1.
64
TABLE 4. 1: Summarize of piecewise linear constraint formation for each equipment
Equip. Operating Set of piecewise Set belong Eqs.
(quadrant) (name) (elements selected) #
𝐷𝐷𝐺
DDG 1st/2nd 𝐿𝑖 {𝐿1𝑠𝑡 4𝑡ℎ
𝑛 , 𝐿𝑛 }
𝐸𝑆𝑆
ESS full 𝐿𝑖 {𝐿1𝑠𝑡 2𝑛𝑑 3𝑟𝑑 4𝑡ℎ
𝑛 , 𝐿𝑛 , 𝐿 𝑛 𝐿𝑛 }
𝑃𝐶𝐶 1𝑠𝑡 2𝑛𝑑 3𝑟𝑑 4𝑡ℎ
TR full 𝐿𝑖 {𝐿𝑛 , 𝐿𝑛 , 𝐿𝑛 𝐿𝑛 }
𝑏𝑟
LINE full 𝐿𝑖 {𝐿1𝑠𝑡 2𝑛𝑑 3𝑟𝑑 4𝑡ℎ
𝑛 , 𝐿𝑛 , 𝐿 𝑛 𝐿𝑛 }
X Y Z
𝐷𝐷𝐺,3𝛷 𝐷𝐷𝐺,3𝛷
DDG 𝑃𝑖,𝑡 𝑄𝑖,𝑡 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 (4.66)
𝐸𝑆𝑆,3𝛷
ESS 𝑑𝑐ℎ,3𝛷
𝑃𝑖,𝑡 𝑐ℎ,3𝛷
− 𝑃𝑖,𝑡 𝑄𝑖,𝑡 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐸𝑆𝑆 (4.67)
𝑃𝐶𝐶,3𝛷 𝑃𝐶𝐶,3𝛷
TR 𝑃𝑖,𝑡 𝑃𝑖,𝑡 𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝑃𝐶𝐶 (4.68)
LINE 𝑏𝑟
𝑃𝑘,𝛹,𝑡 𝑏𝑟
𝑄𝑘,𝛹,𝑡 𝐼𝑘𝑏𝑟,𝑚𝑎𝑥 𝑉𝑖(𝑘),𝛹,𝑡 ,∀𝑘 ∈ Ω𝐿𝐼𝑁 , ∀𝑖 ∈ Ω𝐵𝑈𝑆 (4.69)
where, 𝑛 = 1,2, … , 𝑛𝑝𝑤 ; 𝛹 ∈ ℱ, ∀𝑡 ∈ 𝑁𝑇
4.3.2 MILP Model
The proposed MILP model can be formulated by the set of linearized constraints and objective
function as following:
min ∑𝑡∈Ω𝐷𝐷𝐺[(𝐶𝑡𝑃𝐶𝐶 + 𝐶𝑡𝐷𝐷𝐺 )(1 − 𝑏𝑡𝑚 ) + 𝐶𝑡𝐶𝑈𝑅𝑇 𝑏𝑡𝑚 + 𝐶𝑡𝑉𝐼,𝐿𝑖𝑛 ] (4.70)

𝑿𝑀𝐼𝐿𝑃,3𝛷
subject to:
• Voltage and load curtailment constraints: (4.6), (4.7), (4.44), (4.45), (4.46), (4.9).
• Power flow constraints: (4.52), (4.53), (4.56), (4.57), (4.69).
• DDG operation constrains: (4.19), (4.20), (4.21), (4.22), (4.23), (4.26), (4.27), (4.28), (4.66).
• ESS operational constraints: (4.30), (4.31), (4.32), (4.67), (4.34), (4.35), (4.36).
• Power exchange constraints: (4.37), (4.38), (4.68), (4.40), (4.41), (4.42).
The details of the decision variables of the proposed model are also summarized in Appendix F.
65
4.4 Numerical Results
In this section, the well-known balanced 33-bus and unbalanced 25-bus test system have been chosen
for testing the proposed MINLP and MILP model. The characteristic of these test system can be
found in [1]. The planning horizon is set to 24 hours representing a typical day. The daily market
electricity price and the percentage of annual peak load are taken from NYISO [89] and [90]. The
wind speed and solar irradiance data are given in Table G. 2 of Appendix G. The load curtailment
cost and the penalty coefficient of the voltage violation are obtained from [91] and [92]. The
characteristic of the DDG, WTs, PVs, ESS, transformer, and transmission line are summarized in the
Table G. 4 and Table G. 5, respectively. The weighing coefficient of the load at 𝑖 𝑡ℎ bus and its
maximum curtailment duration are assumed to be proportional to their annual peak and are
summarized in Table G. 3 of Appendix G. The parameters of the adopted piecewise linearization
1𝑠𝑡 𝑡ℎ
4 𝑠𝑡 𝑡ℎ
technique is set to 𝜑𝑠𝑒𝑙 = 𝜑𝑠𝑒𝑙 = 90° and 𝑛1𝑝𝑤 = 𝑛𝑝𝑤
4
=15 pieces which is high enough to obtain an
acceptable accuracy [88]. The maximum and minimum voltage of the buses is set to 1.05 and 0.95
per units, respectively.
To study the effectiveness of the proposed ACOPF model considering the microgrid operation in both
grids connected (GC) and islanding (IS), three scenarios (SCEN 1, SCEN 2 and SCEN 3) are
considered. First, in SCEN 1 the microgrid is assumed to operates in grid connected mode all along
the considered time horizon. Then, SCEN 2 assume that the microgrid operates only in islanding
mode. Finally, SCEN 3 assume that the microgrid is disconnected from the substation by opening the
switch ISW (i.e., 𝑏𝑡𝑚 = 1) between 8:00 and 14:00. In this work, the proposed MILP model and the
reference MINLP model have been built in MATLAB environment and solved using “Intlingprog”
and “Fmincon” solvers, respectively. All the simulation are implemented on a PC computer equipped
with 3.40 GHz Quad-Core Intel(R) Core™ i7-6700 CPU with 32GB of RAM.
4.4.1 Optimum Operation of the Microgrid
The MG optimal operation obtained by solving the proposed MILP model for the three
aforementioned scenarios (i.e., SCEN 1, SCEN 2 and SCEN 3) is presented. Figure 4. 6 shows the
total three phase active power for the generation, storage and load for the planning horizon.
Table 4. 2 shows a comparison between SCEN 1, SCEN 2 and SCEN 3 in terms of costs and voltage
violation.
66
FIGURE 4. 6: Total active power profile for the 25-bus test system: (a) SCEN 1 (b) SCEN 2 (c)
SCEN 3
TABLE 4. 2: Optimization results obtained by MILP for 33-bus and 25-bus in difference
scenarios.
Optimization goals 33-bus (balanced) 25-bus (unbalanced)
SCEN 1 SCEN 2 SCEN 3 SCEN 1 SCEN 2 SCEN 3
Total CPCC ($/24h) 1123.22 0.00 805.45 993.63 0.00 706.89
Total CDDG ($/24h) 0.00 5109.92 1676.71 0.00 4802.60 1603.03
Total CCURT ($/24h) 0.00 149.94 145.35 0.00 829.60 414.80
Total CVD ($/24h) 401.80 94.95 301.81 158.24 56.99 124.32
Total cost ($/24h) 1525.03 5354.81 2929.32 1151.87 5689.20 2849.04
min{𝑉𝑖 ,𝛹,𝑡 } (p.u.) 0.9716 1.0124 0.9719 0.9753 1.0131 0.9753
Avg. Runtime (s) 4.85 58.78 34.34 23.89 63.57 68.04
Due to the availability of the main grid during SCEN 1, the demand can be fully supplied by the grid
and the resources located in the test system. Thus, there is no load curtailment in this scenario
compared to other two scenarios. On the other hand, 0.38% of energy demand and 0.19% of energy
demand has to be respectively curtailed for the scenarios SCEN 2 and SCEN 3, not only for balancing
the power in the system but also for satisfying the secondary control of the DDG units such as voltage
restoration layer. The cost associated with the power exchange with the main grid is zero in SCEN 2
as the microgrid is isolated from the main grid during this scenario. The PV and WT renewable
67
resources account for 3.91% of the total generated energy for the three scenarios. In the grid connected
mode, the DDG units are turned off due to the high cost of fuel consumption compared to the cost of
energy imported from the main grid. Thus, no active power is generated by the DDG units and its
reactive power is accordingly bounded to zero by (4.28). In other words, as it is not economical to
generate active power by DDGs during grid connected operation mode, their reactive power
generation will accordingly be bounded to zero by (4.28). Hence, no reactive power is generated. As
the DDG units are generating no reactive power in this case, the cost of voltage deviation during the
SCEN 1 is high. For the same reason, the cost of voltage deviation in the SCEN 2 and SCEN 3 is
lower than that in SCEN 1. Nevertheless, the total operation cost during SCEN 1 is always lower than
that in SCEN 2 and SCEN 3. That is why the IS mode is usually operated during emergencies and/or
when the main grid is not available. In term of the minimum voltage, they are limited the minimum
voltage magnitude constraints.
4.4.2 Accuracy Evaluation
For evaluating the accuracy, the results obtained by the proposed MILP model are compared to those
obtained by the original MINLP model. For the sake of the comparison with the solutions obtained
by the proposed MILP, the MINLP model is changed to a nonlinear problem (NLP) by setting all
binary variables according to those obtained by solving the proposed MILP model [26]. The third
scenario (SCEN 3) is used and the solution obtained via NLP and MILP as well as the relative error
obtained by the solving the models are provided in Table 4. 3.
TABLE 4. 3: Comparison of results obtained by MILP to those obtained by NPL

Error indices 33-bus (balanced) 25-bus (unbalanced)
NLP MILP Error NLP MILP Error
Total energy DDG (MWh) 23.094 22.858 1.02% 20.721 20.548 0.84%
Total energy curtailment (kWh) 44.1 44.1 0.00% 122 122 0%
Total energy losses (MWh) 1.907 1.930 1.20% 1.632 1.676 2.68%
min{𝑉𝑖 ,𝛹,𝑡 } (p.u.) 0.9703 0.972 0.17% 0.9790 0.9753 0.38%
min{𝛿𝑖 ,𝛹,𝑡 } (deg.) -0.50 -0.496 0.14% -120.66 -120.87 0.17%
max{𝛿𝑖 ,𝛹,𝑡 } (deg.) 90.418 90.446 0.03% 211.31 211.36 0.02%
𝑏𝑟
max{|𝐼𝑖 ,𝛹,𝑡 |} (kA) 0.322 0.321 0.27% 0.640 0.643 0.35%
Avg. Runtime (min) 137.22 0.572 612.01 1.134
Table 4. 3 shows that the relative errors in the calculation of the total energy produced by the DDG
units are around 1%. In terms of power loss, the maximum absolute error in the lines active power
losses by lines is lower than 3 × 10−3 per unit. Nonetheless, the error is accumulated when
68
calculating the total energy losses along the considered time horizon and lead to 1.20% and 2.68%
for 33-bus and 25-bus test systems, respectively.
On the other hand, the relative error of the voltage magnitude, the voltage phase angle, and the current
is lower than 0.5% which validates the accuracy of the proposed MILP model. Figure 4. 7 and
Figure 4. 8 show the voltage profile at time t = 7: 00 (the MG operate in GC mode) and t = 8: 00
(the MG operate in IS mode) obtained by proposed MILP and NPL for the 33-bus and 25-bus test
systems, respectively. As Table 4. 3 shows, the proposed MILP is computationally superior compared
to NLP which is a crucial for real-time operation and stochastic planning applications.
bus test case obtained by NPL and MILP
bus test case obtained by NPL and MILP: (a) Phase A, (b) Phase B, (c) Phase C
69
4.4.3 Primary Droop and Secondary Control Impact Evaluation
As discussed in the literature, the secondary control has been adopted for compensating the steady-
state error derived by the primary droop control layer and to restore the microgrid’s voltage and
frequency [16]. In this subsection, the impact of considering the secondary control layer in the OPF
formulation on the microgrid’s operation is evaluated. To this end, the constraints (4.19)-(4.23) of the
proposed MILP model representing the operation of DDGs with secondary control are modified and
transformed into the well-known droop-based power generation model as expressed in (4.71)-(4.75)
[54], i.e., only primary control. Then, the set of decision variables, included in Table F. 1 of Appendix
𝑣
F, are modified by eliminating a set of 𝑢𝑖,𝛹,𝑡 and introducing the system frequency 𝜔𝑡 as variables to
be solved [54].
𝐷𝐷𝐺,3𝛷 𝑝 𝐷𝐷𝐺,3𝛷
𝑃𝑖,𝑡 = [(ω∗ − 𝜔𝑡 )/𝑚𝑖 ]𝑏𝑡𝑚 + (1 − 𝑏𝑡𝑚 )𝑃𝑖,𝑡 /3 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 (4.71)
𝐷𝐷𝐺,3𝛷 𝑞 𝐷𝐷𝐺,3𝛷
𝑄𝑖,𝑡 = [(𝑉𝑖∗ − 𝑉𝑖,𝑡
𝑎
)⁄𝑛𝑖 ]𝑏𝑡𝑚 +(1 − 𝑏𝑡𝑚 )𝑄𝑖,𝑡 /3 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ,∀𝑡 ∈ 𝛺𝑇 (4.72)
𝐷𝐷𝐺,3𝛷 𝐷𝐷𝐺
𝑃𝑖,𝑡 = ∑𝛹∈ℱ 𝑃𝑖,𝛹,𝑡 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 (4.73)
𝐷𝐷𝐺,3𝛷 𝐷𝐷𝐺
𝑄𝑖,𝑡 = ∑𝛹∈ℱ 𝑄𝑖,𝛹,𝑡 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 (4.74)
𝑎 𝑚 𝑏 𝑚 𝑐 𝑚
𝑉𝑖,𝑡 𝑏𝑡 = 𝑉𝑖,𝑡 𝑏𝑡 = 𝑉𝑖,𝑡 𝑏𝑡 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 , (4.75)
Finally, both MILP OPF models, with and without the secondary control, are used to solve the 33-
bus test system considering SCEN 2. The results are derived and the effect of primary droop and
secondary controls are evaluated accordingly. For the ease of notation, we denote the proposed MILP
model in which both primary and secondary control layer have been considered as MILP2nd .
Similarly, we denote the modified model with only droop primary control layer as MILPdroop.
First, with the application of secondary control layer the voltage magnitudes at the buses where the
DDG units are connected are restored. Figure 4. 9 shows the voltage magnitude profile for bus #25
where a DDG is connected.
In SCEN 2, the microgrid is operated in IS mode along the planning horizon 24h. As can be seen in
Figure 4. 9, with MILP2nd the voltage restoration layer of secondary control is activated and the
voltage magnitude at the bus #25 is restored to the secondary voltage control reference (𝑉 ∗∗ =
70
1.05 𝑝. 𝑢.). Nonetheless, the result derived by the MILP droop, show that the voltage magnitude at bus
#25 follows the primary control voltage droop with a reference 𝑉 ∗ = 1.01 𝑝. 𝑢..
FIGURE 4. 9: Daily voltage magnitude profile of bus #25 obtained by the MILP2nd and
MILPdroop for the 33-bus test system SCEN 2
Additionally, Figure 4. 10 shows the voltage profile for all the system buses during the at 12:00 am
in SCEN 2.
FIGURE 4. 10: Voltage magnitude profile with and without secondary control for the 33-bus
test system for SCEN 2 at 12:00 am
As can be seen in Figure 4. 10, the voltage magnitude profile derived by the MILP2nd follows 𝑉 ∗∗ at
the DDG units connection points, while the voltage profile obtained using the MILP droop model is
always maintained below 𝑉 ∗ by following the primary droop control. Figure 4. 11 shows the power
flowing in the branch #24 i.e., |𝑃24 |. This branch is connected between bus #24 and #25. As can see,
the power flows obtained by MILPdroop are different compared to those obtained by MILP2nd .
FIGURE 4. 11: Daily power flowing on branch #24 obtained by MILP2nd and MILPdroop for the
33-bus test system for SCEN 2
71
The secondary control has also been used for maintaining the system frequency to its nominal value
and make the microgrid ready for synchronization and reconnection to the main grid. The daily
frequency of the microgrid system is shown in Figure 4. 12.
The system frequency derived by the proposed MILP2nd model can be directly specified by 𝜔𝑡 =
1.00 𝑝. 𝑢. or e.g., 60 𝐻𝑧 in this work. Nonetheless, in the MILPdroop model with only the primary
droop control adoption, the system frequency is a variable to be found [54]. In this case, this frequency
is dependent on the active droop characteristic 𝑃 − 𝜔 and the normal angular frequency, set to 𝜔∗ =
1.00 𝑝. 𝑢. in this work.
FIGURE 4. 12: Daily system frequency obtained by the MILP2nd and MILPdroop for the 33-bus
test system SCEN 2
Given that the results of the MILP2nd closely match those of the detailed MINLP model as depicted
in Table 4. 3, the results of the comparison between MILP2nd OPF model and MILPdroop OPF model
demonstrate that the MILP droop OPF formulation is not sufficient for representing the actual behavior
of the microgrid with hierarchal control i.e., when a secondary control layer is employed.
4.4.4 Evaluation of Load Shedding Switching State Cycle
In this section, in order to evaluate the impacts of constraints (4.44)-(4.46) employed to limit the
number of load shedding cycles on the microgrid operation, we consider a case in which it is assumed
that the DDGs #3 to #5 are disactivated. This will force a large portion of load to be shed in order to
balance the generation and demand. Figure 4. 13 shows the total number of switching cycles for each
load 𝑖 within 24 hours for 33-bus test system in SCEN 2 with DDGs #3 to #5 disactivated.
As can be seen in Figure 4. 13, the maximum number of switching cycles for the case without
considering (4.44)-(4.46) are 5.5 cycles. It this necessary to note that, 0.5 cycle represents a single
switching state, e.g., on to off or vice versa. Two more cases where the number of cycles is limited
by 𝑁 𝑐𝑢𝑟𝑡 = 1 and 𝑁 𝑐𝑢𝑟𝑡 = 3 are also provided. In these cases, the number of cycles has been limited
to 𝑁 𝑐𝑢𝑟𝑡 = 1 and 𝑁 𝑐𝑢𝑟𝑡 = 3 cycles.
72
FIGURE 4. 13: Number of load shedding switching state cycle
The loads energy demand has been shed from the network by 26.95%, 27.27%, and 26.47% for the
𝑐𝑢𝑟𝑡
cases without (4.44)-(4.46), with 𝑁 𝑐𝑢𝑟𝑡 = 1, and with 𝑁 𝑐𝑢𝑟𝑡 = 3, respectively. The 𝑏19,𝑡 for these
cases is shown in Figure 4. 14. As can be seen, the load in bus #19 is more often switched between
the shedding and supply states in the “no limit” case compared to other cases. This situation might
not be satisfying for the consumers/operators as there is no control over the load shedding cycles.
Nonetheless, with (4.44)-(4.46) applied, the number of cycles is limited.
FIGURE 4. 14: Profile of load shedding at bus #19
4.4.5 Evaluation of ESS with primary droop and secondary control

4.4.5.1 Formulation for ESS with primary droop and secondary control
With the power electronics inverter interface, the ESS has the ability to be operated in droop control
mode with a supervisory secondary control layer. In this mode, the ESS units participate with the
dispatchable resources microgrid to maintain the system frequency and the voltage to the nominal
value. The ESS operates in primary droop control mode with a supervisory secondary control layer
when it discharges, while during charging mode they act like a load and are as such operated in the
PQ mode. For considering the ESS operation in primary droop control mode with a supervisory
73
secondary control layer, a new set of constraints expressed by (4.76)-(4.80) are added to the
constraints (4.30)-(4.36), representing the operation of the ESS in PQ mode:
𝑝
𝑀𝐺
[− 𝑘𝑖 𝜔 (𝛿𝛹,𝑡 − 𝜋⁄2)⁄𝑚𝑖 ]𝑏𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚
𝑖,𝑡 𝑏𝑡 𝑏𝑖 , 𝛹=𝑎
𝑑𝑐ℎ 𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚 𝑝
𝑃𝑖,𝛹,𝑡 𝑏𝑖,𝑡 𝑏𝑡 𝑏𝑖 𝑀𝐺
= { [− 𝑘𝑖 𝜔 (𝛿𝛹,𝑡 + 𝜋⁄6)⁄𝑚𝑖 ]𝑏𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚
𝑖,𝑡 𝑏𝑡 𝑏𝑖 , 𝛹=𝑏
𝑝
𝑀𝐺
[− 𝑘𝑖 𝜔 (𝛿𝛹,𝑡 − 7𝜋⁄6)⁄𝑚𝑖 ]𝑏𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚
𝑖,𝑡 𝑏𝑡 𝑏𝑖 , 𝛹=𝑐
,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝑡 ∈ 𝛺𝑇 (4.76)
𝐸𝑆𝑆 𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚 𝑞

𝑄𝑖,𝛹,𝑡 𝑏𝑖,𝑡 𝑏𝑡 𝑏𝑖 𝑣,𝑖𝑛𝑡
= [(𝑉𝑖∗ − 𝑉𝑖,𝛹,𝑡 + 𝑢𝑖,𝛹,𝑡 )⁄𝑛𝑖 ]𝑏𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚
𝑖,𝑡 𝑏𝑡 𝑏𝑖
,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝛹 ∈ ℱ,∀𝑡 ∈ 𝛺𝑇 (4.77)
𝑑𝑐ℎ,3𝛷 𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚

𝑃𝑖,𝑡 𝑏𝑖,𝑡 𝑏𝑡 𝑏𝑖 = 𝑏𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚 ∑
𝑖,𝑡 𝑏𝑡 𝑏𝑖
𝑑𝑐ℎ
𝛹∈ℱ 𝑃𝑖,𝛹,𝑡 ,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝑡 ∈ 𝛺𝑇 (4.78)
𝐸𝑆𝑆,3𝛷 𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚

𝑄𝑖,𝑡 𝑏𝑖,𝑡 𝑏𝑡 𝑏𝑖 = 𝑏𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚
𝑖,𝑡 𝑏𝑖,𝑡 𝑏𝑖
𝐸𝑆𝑆
∑𝛹∈ℱ 𝑄𝑖,𝛹,𝑡 ,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝑡 ∈ 𝛺𝑇 (4.79)
𝑉𝑖,𝛹,𝑡 𝑏𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚

𝑖,𝑡 𝑏𝑡 𝑏𝑖 = 𝑉𝑖∗∗ 𝑏𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚
𝑖,𝑡 𝑏𝑡 𝑏𝑖 ,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.80)
These set of constraints are derived from those implemented for the DDG units operating with
secondary control. Nonetheless, 𝑏𝐸𝑆𝑆
𝑖,𝑡 introduced in (4.76)-(4.80) to ensure that the constraints are
activated only when the ESS is in discharging state i.e., 𝑏𝐸𝑆𝑆

𝑖,𝑡 = 1. Moreover, the binary input
parameter 𝑏𝑡𝑚 introduced in this set of constraints guaranties that the ESS operates in PQ mode during
grid-connected operation mode. The binary input parameter 𝑏𝐸𝑆𝑆𝑚
𝑖 represents the operation mode for
the ESS and is set to “0” for PQ mode and “1” for droop control mode with a supervisory secondary
control layer. The active and reactive droop gains of the ESS units are determined using the capacity-
based method similar to those used for the DDG unit. As can be seen in (4.76)-(4.80), there are
nonlinear terms in this set of constraints due to the multiplication between continuous variables and
the binary variable 𝑏𝐸𝑆𝑆
𝑖,𝑡 . In order to resolve this nonlinearity, the linearization approach for the
multiplication between continuous and binary variables introduced in Appendix E is adopted. Finally,
𝑑𝑐ℎ 𝐸𝑆𝑆
as 𝑃𝑖,𝛹,𝑡 and 𝑄𝑖,𝛹,𝑡 are introduced for each phase, the power flow constraints are needed to be modified
from (4.10) and (4.11) to (4.81) and (4.82), while respecting to 𝑏𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚
𝑖,𝑡 , 𝑏𝑡 and 𝑏𝑖 as follows:
74
𝛹𝛽
𝐺𝑖𝑗 cos(𝛿𝑖,𝛹,𝑡 − 𝛿𝑖,𝛽,𝑡 )
𝑃𝑖,𝛹,𝑡 = 𝑉𝑖,𝛹,𝑡 ∑𝑗∈Ω𝐵𝑈𝑆 ∑𝛽∈ℱ 𝑉𝑖,𝛹,𝑡 ( 𝛹𝛽 )=
+𝐵𝑖𝑗 sin(𝛿𝑖,𝛹,𝑡 − 𝛿𝑖,𝛽,𝑡 )
𝑃𝐶𝐶 𝐷𝐷𝐺 𝑑𝑐ℎ 𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚
𝑃𝑖,𝛹,𝑡 + 𝑃𝑖,𝛹,𝑡 + 𝑃𝑖,𝛹,𝑡 𝑏𝑖,𝑡 𝑏𝑡 𝑏𝑖
𝑊𝑇,3𝛷 𝑃𝑉,3𝛷
𝑃𝑖,𝑡 + 𝑃𝑖,𝑡
+ ( 𝑑𝑐ℎ,3𝛷 𝑐ℎ,3𝛷 )⁄3
,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.81)
+𝑃𝑖,𝑡 (1 − 𝑏𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚
𝑖,𝑡 𝑏𝑡 𝑏𝑖 ) − 𝜂𝐸𝑆𝑆𝑖 𝑃𝑖,𝑡
𝐿
( −𝑃𝑖,𝛹,𝑡 (1 − 𝑏𝑡𝑐𝑢𝑟𝑡 ) )
𝛹𝛽
−𝐵𝑖𝑗 cos(𝛿𝑖,𝛹,𝑡 − 𝛿𝑖,𝛽,𝑡 )
𝑄𝑖,𝛹,𝑡 = 𝑉𝑖,𝛹,𝑡 ∑𝑗∈Ω𝐵𝑈𝑆 ∑𝛽∈ℱ 𝑉𝑖,𝛹,𝑡 ( 𝛹𝛽 )=
+𝐺𝑖𝑗 sin(𝛿𝑖,𝛹,𝑡 − 𝛿𝑖,𝛽,𝑡 )
𝑃𝐶𝐶 𝐷𝐷𝐺 𝐸𝑆𝑆 𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚
+𝑄𝑖,𝛹,𝑡 + 𝑄𝑖,𝛹,𝑡 + 𝑄𝑖,𝛹,𝑡 𝑏𝑖,𝑡 𝑏𝑡 𝑏𝑖
𝑊𝑇,3𝛷
𝑄𝑖,𝑡
+ ( 𝐸𝑆𝑆,3𝛷 )⁄3 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝛹 ∈ ℱ, ∀𝑡 ∈ 𝛺𝑇 (4.82)
+𝑄𝑖,𝑡 (1 − 𝑏𝐸𝑆𝑆 𝑚 𝐸𝑆𝑆𝑚
𝑖,𝑡 𝑏𝑡 𝑏𝑖 )
𝐿
( −𝑄𝑖,𝛹,𝑡 (1 − 𝑏𝑡𝑐𝑢𝑟𝑡 ) )
4.4.5.2 Evaluation of ESS with primary droop and secondary control
In this subsection, the operation of the ESS in primary droop control mode with a supervisory
secondary control layer is evaluated. A new ESS is added into microgrid with the capability to operate
in both the power control mode or in primary droop control mode with a supervisory secondary
control layer. As such, two new case studies, CASE P and CASE2nd , are introduced for the proposed
MILP model. In CASE P, the ESS is operated power control PQ mode while in CASE 2nd it is operated
in droop control mode with a supervisory secondary control layer. Both case studies are used to solve
the 33-bus test system considering SCEN2. The ESS is allocated to bus #13 of the test system with
the following characteristic: 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 = 1.2 MVA, 𝐸𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 = 4.5 MWh, 𝐸𝑖𝐸𝑆𝑆,𝑚𝑖𝑛 = 0.9 MWh, and
𝜂𝑖𝐸𝑆𝑆 = 95%.
Figure 4. 15 shows the charging/discharge state of the new ESS allocated at bus #13 as well as the
voltage for this bus. As can see at a particular period of time i.e., 12h00-13h00, the ESS operated in
discharging mode for both cases i.e., the ESS with and without the primary droop control mode with
a supervisory secondary control layer. As expected, the ESS operating in droop control mode with a
supervisory secondary control layer can maintain the voltage refers to their nominal voltage setting
𝑉 ∗∗ . Nonetheless, with ESS in PQ mode the voltage for the bus #13 in this case is below 𝑉 ∗∗ of the
DDG units. For the rest of period, the ESS for both cases were operated similarly.
75
FIGURE 4. 15: Daily profile for (a) binary indicate charging/discharging state of ESS localized
at bus #13 (b) voltage for the bus #13
Figure 4. 16 shows the active power profile for a particular period of ESS operating in discharging
mode i.e., 12h00-13h00. When the ESS operates in PQ mode, the active power has been scheduled
to inject its power into the network according to the optimal operation of the microgrid. Nonetheless,
when the ESS operates in droop control mode with a supervisory secondary control layer, the ESS
𝑑𝑐ℎ 𝑑𝑐ℎ
injects the specific amount of active power by 𝑃13,𝐴,𝑡=12 = 566.81 kW and 𝑃13,𝐴,𝑡=13 = 559.96 kW
𝑀𝐺 𝑀𝐺
corresponding to the droop settings under 𝛿𝐴,𝑡=12 = 89.932° and 𝛿𝐴,𝑡=13 = 89.933°, respectively.
FIGURE 4. 16: Active discharged power of ESS localized on bus #13 for both case ESS with
and without considering droop+2nd control
The overall optimization results have been summarized in Table 4. 4. The results show that the total
cost is not dependent on the operating mode of the ESS. For the same reason, no load is curtailed in
both cases. Nonetheless, the minimum voltage for CASE 2nd is slightly higher than that in CASE P due
to primary droop and the secondary control adopted. Finally, the maximum branch current and total
energy loss were higher in CASE 2nd compared to those obtained in CASEP .
76
TABLE 4. 4: Optimization results for CASE2nd and CASEP
Indicators 𝐂𝐀𝐒𝐄𝟐𝐧𝐝 𝐂𝐀𝐒𝐄𝐏
Total CDDG ($/24h) 5113.15 5113.15
CURT
Total C ($/24h) 0.00 0.00
Total cost ($/24h) 5188.24 5188.24
min{𝑉𝑖,𝛹=𝐴,𝑡 } (p.u.) 1.014 1.013
max{𝑉𝑖,𝛹=𝐴,𝑡 } (p.u.) 1.05 1.05
min{𝛿𝑖,𝛹=𝐴,𝑡 } (deg.) 88.000 88.113
max{𝛿𝑖,𝛹=𝐴,𝑡 } (deg.) 90.779 90.609
𝑏𝑟
max{𝐼𝑘,𝛹=𝐴,𝑡 } (kA) 0.102 0.099
Total energy loss (MWh) 0.912 0.896
4.5 Conclusion
An MINLP model for the optimal operation problem of unbalanced microgrid with hierarchical
control structure was proposed. The model considers the microgrid philosophy of operation during
GC and IS in a single comprehensive model. The energy storage system and the intermittent resources
such as WTs and PVs were considered in the proposed model. Additionally, the proposed MINLP
model was linearized and the MILP model of the problem was obtained which can be solved
efficiently with commercial solvers. The obtained solutions can guarantee the optimal operation of
the microgrid respecting to their philosophy of operation during both operation mode. Numerical
simulation studies were conducted to prove the effectiveness of the proposed models. The error
assessment was also conducted for validating the proposed model. The results showed that the
proposed MILP model achieves the optimum results with significantly lower computational time
compared to that in the MINLP. Such performance is crucially needed in the real-time and stochastic
planning application for the microgrid.
77
Chapter 5 Resilience Oriented Microgrid Planning
Y. Vilaisarn, Y. R. Rodrigues, and M. Abdelaziz, “A Deep Learning based Multi-Objective
Optimization for the Planning of Resilience Oriented Microgrids in Active Distribution System”.
Submitted in IEEE Access, Mars 2022.
Résumé
Lorsqu'il est confronté à des phénomènes météorologiques violents, un système de distribution peut
subir la perte ou la défaillance d'un ou plusieurs de ses composants, ce que l'on appelle les systèmes
contingences à niveau N-K. Néanmoins, en tirant parti des sectionneurs électriques et de la
disponibilité croissante des ressources énergétiques distribuées, un système de distribution peut être
formé en micro-réseaux pour résister à de telles éventualités avec une interruption minimale de
l'alimentation. Dans cette perspective, ce travail propose un nouveau cadre d'optimisation à deux
niveaux pour la planification de micro-réseaux dans un système de distribution actif dans une
perspective de résilience. Pour cela, tout d'abord, le niveau supérieur traite un problème d'optimisation
multi-objectif pour allouer, de manière optimale les ressources d’énergies distribuées et les
sectionneurs électriques dans le réseau de distribution en équilibrant les autres objectifs que sont le
coût,la résilience et l'impact environnemental. Ensuite, le niveau interne traite du fonctionnement
optimal des micro-réseaux qui peuvent être créés en exploitant les ressources énergies distribuées et
les sectionneurs électrique d'isolement alloués dans le niveau supérieur. Ce cadre utilise des modèles
d'apprentissage de réseaux de neurones profonds pour la prise en compte du comportement
opérationnel des micro-réseaux, en surmontant les simplifications de modélisation visant à alléger la
charge de calcul. En effet, cette charge de calcul peut compromettre l’étude des solutions de
planification. Il faut prendre en compte le contrôle hiérarchique, les charges et les générations
stochastiques, ainsi que la nature stochastique des pannes de ligne induites par la météo, qui sont des
aspects particulièrement critiques dans le cadre d'une planification axée sur la résilience. Des études
de cas de simulation sont développées pour démontrer l'efficacité de la méthode de planification
proposée.
Abstract
When facing severe weather events, a distribution system may suffer from the loss or failure of one
or more of its components, the so-called N-K contingencies. Nevertheless, taking advantage of the
system’s isolate switches and the increasing availability of distributed energy resources (DERs), a
distribution system may be clustered into microgrids able to withstand such contingencies with
78
minimal power interruption. In this perspective, this work proposes a novel bilevel optimization
framework for planning microgrids in active distribution systems under a resilience-oriented
perspective. For this, first, the outer level optimization features a multi-objective problem seeking to
optimally allocate DERs and isolate switches in the distribution network while balancing the
competing objectives of cost, resilience, and environmental impact. Next, the inner level handles the
optimization problem pertaining to the optimal operation of the microgrids that can be created by
harnessing local DERs and isolate switches allocated in the outer level. Further, given the proposed
approach resilience-oriented focus, the developed framework employes deep learning models based
on deep neural network (DNN) architectures trained using Bayesian Regularization Backpropagation
(BRB) technique. This strategy allows for avoiding the modeling simplifications typically employed
to alleviate the computational burden that can otherwise jeopardize planning solutions' feasibility.
Thus, enabling the accurate consideration of microgrids' operational behavior, including hierarchal
controls and the stochastic nature of loads, generation, and weather-induced line failures, especially
critical aspects under resilience-oriented planning. Simulation case studies are developed to
demonstrate the effectiveness of the developed planning framework.
5.1 Introduction
Regarding to the challenges discussed in the literature in chapter 1, in this chapter we propose a
general framework for resilience oriented microgrid(s) planning that exploit deep learning to alleviate
the computational burden associated with planning problem without sacrificing the model accuracy.
The remainder of this chapter is organized as follows: First, basic outline for the proposed resilience
oriented microgrid planning framework is presented in Section 5.2. Subsequently, the problem
formulation for the outer and inner levels are descripted in Section 5.3. Then, the data preparation
stage for the proposed planning approach is presented in Section 5.4. Next, a deep learning approach
for a stochastic optimal operation level are discussed in Section 5.5. Later, Section 5.6 exhibits the
detail for optimal microgrid planning approach NSGA-II based DNN models. Finally, section 5.7 and
5.8 present the results and concludes the paper, respectively.
5.2 Proposed Resilience Oriented Planning Framework

The availability of isolate switches (ISWs) along with the locations and capacities of the DERs,
including DG units, renewable DG units and energy storage (ES) units, are salient for the successful
creation of MG(s) in a distribution network. Additionally, the optimal operation of the MG’s
resources should be considered to ensure the MG successful operation, minimize its operational costs
and maximize its resilience. As such, the proposed resilience-oriented planning framework seeks to
79
identify the optimal allocation of DERs and ISWs in a distribution network considering the optimal
operation of the MG(s) that will be created using these components. To this end, the analyzed problem
is formulated as a stochastic bilevel optimization framework. First, an outer level featuring a
metaheuristic multi-objective optimization is adopted for the optimal allocation of the DERs and
ISWs seeking to balance the competing objectives of cost, resilience and environmental impact. Then,
an inner level is proposed to handle the optimization problem pertaining to the optimal operation of
the MG(s) that can be created by the DERs and ISWs allocated in the outer level. The problem of the
MG’s optimal operation is cast as a linear optimum power flow (LOPF) problem. In this work, the
adopted LOPF model considers the operational behavior of the MG in both grid-connected (GC) and
islanded (IS) modes of operation. Additionally, different scenarios are considered to account for the
stochastic nature of the HILP events, as well as renewable DG units. In this sense, despite using a
LOPF model, the detailed modeling of MGs and consideration of different stochastic scenarios in the
inner level along with the metaheuristic nature of the outer level make solving the LOPF, for each of
the stochastic scenarios for each individual in the metaheuristic optimization’s population, using a
numerical optimization solver computationally challenging. In other words, the combination of the
need for numerous iterations to solve the LOPF of the inner level and the large populations from the
metaheuristic presented in the outer level before reaching the final planning solution, make this
problem computationally extremely demanding that it may be unsolvable with the traditional
analytical methods. In this sense, in this work, a deep learning technique is adopted for deriving the
information required from the LOPF solution for the stochastic scenarios under consideration. With
this technique, DNN models are developed and trained to predict the solution delivered by the inner
level that comprises the LOPF solution with the numerous stochastic scenarios, significantly reducing
the computation burden, and thus eliminating the possible requirement for modelling simplication.
This new perspective provides an effective strategy for solving MG planning framework, overcoming
computation constraints that limit the consideration of detailed systems' modeling and scenarios,
which are especially critical features for guaranteeing feasible solutions under a resilience-oriented
outlook.
The full outline of the proposed resilience-oriented MG planning framework is shown in Figure 5. 1.
The proposed planning framework comprises four cascading stages for delivering the final optimal
allocation of the DERs and switches. 1) The first stage is concerned with data preparation. In this
stage, the distribution system under consideration and its planning elements are identified along with
the uncertainty scenarios describing the stochastic nature of the system and the planning elements.
The target system for this work is a typical distribution system with the ability to disconnect from the
80
START
Stage 1: Data preparation

Planning elements Initial test system Scenarios generation
Generate wind, solar
Dispatchable DG
Microgrid and load scenarios
ISWsub
Wind turbine Seasonal Seasonal
Photovoltaic
PCC MG Severe Normal
situation situation
Energy storage system Candidate branch Microgrid PCC scenarios scenarion
Isolate switch Candidate bus Substation
A B
Stage 2: MGs planning problem formulation

Problem formulation for:
C Outer level: optimum allocation
Inner level: optimal operation
1) candidate location and size for resources Stage 4: Optimal planning approach
2) candidate location for switches Problem objectives
q Maximize resilience
Generate set of possible configuration
q Best in economic
Set of samples q Environnent friendly
A B
Outer level: Optimal allocation layer
Generate microgrid formation SUB-START
indication vectors and matrices when stage 1,2 and 3 ready
Initialize variables
Run optimal operation for severe
and normal opeartion scenarios
Microgrid optimal operation Inner level:
C Optimal operation
All samples simulated
Identify outer Evaluate multi-
level LOPF objective functions
dependent indices
Optimize variables
calcalate targets Iter=iter+1
Criteria
Train optimal operation targets satisfied
model based deep
yes
learning technique
DNN models Resilience oriented
microgrid(s)
Stage 3: DNN models determination
END
FIGURE 5. 1: Basic outline for the proposed Resilience Oriented Microgrids Planning
framework
main grid via a main isolate switch, denotated by ISW sub . The planning elements include dispatchable
distributed generation (DDG), wind turbine (WT), photovoltaic (PV), energy storage system (ESS)
81
and isolate switch (ISW). Candidate buses and branches are specified as potential locations for the
system’s planning elements. Stochastic scenarios are considered for allowing for the uncertainty
associated with the MG(s) creation, i.e., HILP, and operation (renewable generation and load). These
uncertainty scenarios are seasonal, and include a set of scenarios representing the normal operation
of the system and another set of scenarios representing the severe HILP events in which N-K
contingencies occurred in the network within a certain time interval. 2) The second stage corresponds
to the problem formulation of the proposed stochastic bilevel optimization framework. In this
formulation, the outer level’s objective functions are cost, resilience and environmental metrics
(indices) that depend on the solution of LOPF for different stochastic scenarios in the inner level. 3)
The third stage seeks to deliver the deep learning models that are used to determine the stochastic
LOPF dependent indices required for the outer level. To this end, a set of samples DERs and switches
allocation is developed considering randomly generated combinations of possible installation
decisions for the candidate planning elements locations and capacities. Next, the stochastic LOPF is
solved using a numerical optimization solver for each of the allocation samples under consideration
for each of the stochastic scenarios. Based on the LOPF solutions obtained for these samples, the
outer level cost, resilience and environmental indices are calculated. The calculated indices are
considered as the target values of the allocation samples. The allocation samples and their
corresponding target values are then employed for training the Deep Neural Network (DNN) models
that map the planning elements installation decisions to the outer level’s cost, resilience and
environmental indices. Finally, in the fourth stage, the optimal planning approach is held. Using a
population-based metaheuristic optimization exploiting the DNN models derived in the third stage,
the optimal solution describing the optimal allocation of DERs and ISWs is determined respecting
the optimal trade-off between the multi-objective functions, i.e., resilience, economic and
environment impact. A flowchart outlining the proposed strategy four main stages is detail depicted
in Figure 5. 1. Each of these four stages is individually addressed in the following sections.
5.3 Stage I: Data Preparation

In this section, the first stage responsible for data preparation is addressed, including the modeling of
distribution system and planning elements, as well as generation uncertainty, normal operation and
HILP event scenarios.
5.3.1 Distribution System and Planning Elements
The distribution system considered in this work is a typical distribution network with the ability to
disconnect from the main grid via a main ISW sub located at the upstream interfacing substation.
82
Candidate buses and branches for the installation of DERs and isolation switches, respectively, are
assumed to be selected arbitrarily by the Distribution System Operator (DSO). The planning elements
include the DDGs, WTs, PVs, ESSs and ISWs. In grid connected mode, the DDGs are controlled to
inject constant amounts of power prespecified by a higher level control layer typically seeking to
realize an economic benefit [3]. On the other hand, in the islanding mode of operation the DDGs are
controlled using droop/2nd control to follow the load variation and maintain the system voltage and
frequency within adequate bounds, allowing for quick synchronization back with main grid when its
ready. The WTs and PVs are intermittent resources and are as such operated in power control mode
to allow for maximum power tracking. The ESSs are interfaced with a power electronic inverter and
are controlled in a power control mode to inject/absorb power from the microgrid as needed. The
ISWs are used to isolate faults in according to the considered N-K contingencies.
5.3.2 WT and PV Generation Model
The power generated by the WTs and PVs depend on the wind velocity and solar irradiance,
respectively, and can be modeled using (5.1) and (5.2) similar to the work in [78].
𝑚
0, 𝑣𝑖,𝑡,𝑠,𝑠𝑒 < 𝑣𝑖𝑐𝑖 , 𝑣𝑖,𝑡,𝑠,𝑠𝑒
𝑚
≥ 𝑣𝑖𝑐𝑜
𝑚
𝑊𝑇 (𝑣𝑖,𝑡,𝑠,𝑠𝑒 −𝑣𝑖𝑐𝑖 )
𝑃𝑖,𝑡,𝑠,𝑠𝑒 = 𝑃𝑖𝑊𝑇,𝑠𝑖𝑧𝑒 , 𝑣𝑖𝑐𝑖 ≤ 𝑣𝑖,𝑡,𝑠,𝑠𝑒
𝑚
≤ 𝑣𝑖𝑟𝑎
(𝑣𝑖𝑟𝑎 −𝑣𝑖𝑐𝑖 )
{ 𝑃𝑖𝑊𝑇,𝑠𝑖𝑧𝑒 , 𝑣𝑖𝑟𝑎 < 𝑣𝑖,𝑡,𝑠,𝑠𝑒

𝑚
< 𝑣𝑖𝑐𝑜
,∀𝑖 ∈ Ω𝑐𝑎𝑛 ′ ′′
𝑊𝑇 , ∀𝑡 ∈ Ω 𝑇 , ∀𝑠 ∈ Ω𝑠 ∪ Ω𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.1)
2
𝑃𝑖𝑃𝑉,𝑠𝑖𝑧𝑒 (𝑟𝑖,𝑡,𝑠,𝑠𝑒
𝑚
) ⁄(𝑅𝑖𝑆𝑇𝐷 − 𝑅𝑖𝐶 ) , 𝑟𝑖,𝑡,𝑠,𝑠𝑒
𝑚
≤ 𝑅𝑖𝐶
𝑃𝑉 2
𝑃𝑖,𝑡,𝑠,𝑠𝑒 = 𝑃𝑖𝑃𝑉,𝑠𝑖𝑧𝑒 (𝑟𝑖,𝑡,𝑠,𝑠𝑒
𝑚
) ⁄𝑅𝑖𝑆𝑇𝐷 , 𝑅 𝐶 < 𝑟𝑖,𝑡,𝑠,𝑠𝑒
𝑚
< 𝑅𝑖𝑆𝑇𝐷
{ 𝑃𝑖𝑃𝑉,𝑠𝑖𝑧𝑒 , 𝑟𝑖,𝑡,𝑠,𝑠𝑒
𝑚
≥ 𝑅𝑖𝑆𝑇𝐷
,∀𝑖 ∈ Ω𝑐𝑎𝑛 ′ ′′
𝑃𝑉 , ∀𝑡 ∈ Ω 𝑇 , ∀𝑠 ∈ Ω𝑠 ∪ Ω𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.2)
In this work, it is assumed that the WTs and PVs are operating with unity power factor. Thus, there
is no reactive power injected by these units leading to (5.1) and (5.2) exclusively representing active
power relations.
5.3.3 Load Model
The varying seasonal behavior of the loads is modeled using the annual peak load pattern [90]. In this
sense, the load at bus 𝑖 ∈ Ω𝐵𝑈𝑆 at time 𝑡 ∈ Ω𝑇 can be calculated as the multiplication between the
peak demand at this bus and the annual peak load factor at time 𝑡 ∈ Ω𝑇 for the season under
83
consideration. The active and reactive power demand for bus 𝑖 at time 𝑡 for scenarios 𝑠 and season
𝑠𝑒 can be calculated using (5.3) and (5.4).
𝐿 𝐴𝑃𝐿𝑡,𝑠,𝑠𝑒 𝐿,𝑝𝑒𝑎𝑘
𝑃𝑖,𝑡,𝑠,𝑠𝑒 =( 100
) × 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ Ω𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.3)
𝐿 𝐴𝑃𝐿𝑡,𝑠,𝑠𝑒 𝐿,𝑝𝑒𝑎𝑘
𝑄𝑖,𝑡,𝑠,𝑠𝑒 =( 100
) × 𝑄𝑖,𝑡,𝑠,𝑠𝑒 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ Ω𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.4)
5.3.4 Operation Scenarios
Given the resilience-based planning perspective of the proposed work, two sets of scenarios have
been considered. First, scenarios representing normal system operations without contingencies are
addressed. Second, scenarios representing severe HILP events, in which N-K contingencies occur
during a certain time interval are tackled. These scenarios generation processes are described in the
following subsections.
5.3.4.1 Normal Operation Scenarios
In the normal operation scenarios, the system is assumed operating in grid connected mode
considering uncertainties associated with the renewable resources. For this, historical data sets
comprising different 24-hours’ time-series of wind speeds and solar irradiance are considered. Next,
based on this data a fast forward reduction approach is employed [93], extracting 𝑁𝑠′ scenarios per
season representing the uncertainty associated with the renewable resources. This step seeks to avoid
the higher computation burden, while keep a good approximation of the system. The wind speed and
solar irradiance in a given scenario 𝑠 can be expressed by (5.5) and (5.6), respectively.
𝑚 𝑚 𝑚
𝒗𝒎
𝒔,𝒔𝒆 = [𝑣1,𝑠,𝑠𝑒 , 𝑣2,𝑠,𝑠𝑒 , … , 𝑣𝑁𝑇 ,𝑠,𝑠𝑒 ] ,∀𝑠 ∈ Ω′𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.5)
𝑚 𝑚 𝑚
𝒓𝒎
𝒔,𝒔𝒆 = [𝑟1,𝑠,𝑠𝑒 , 𝑟2,𝑠,𝑠𝑒 , … , 𝑟𝑁𝑇 ,𝑠,𝑠𝑒 ] ,∀𝑠 ∈ Ω′𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.6)
5.3.4.2 Severe HILP Operation Scenarios
For scenarios representing the operation of the distribution system under severe HILP events, a
similar approach to the work developed in [94] is employed. In this sense, by using the component
availability matrix 𝐀𝐕, the availability of the different system components (e.g., branch) at different
time segments in the planning horizon is depicted. This matrix allows for the identification of the
contingency level, as well as the fault location(s) at any instant in the planning horizons. This in turn
enables the development of effective microgrid formation solutions that can meaningfully improve
84
distribution system resiliency. For considering the distribution system operation under severe HILP
events, the following assumptions are made in this work:
• Severe wind events are considered as the main source of contingencies affecting the
distribution network, similar to the work developed in [94].
• Only overhead lines are considered to be vulnerable to severe winds. Distribution substations
and underground lines are able to withstand severe winds as depicted in [95].
• The distribution system buses are assumed to be geographically close enough that the
trajectory of the wind passing through the distribution network cause the entire network to
be exposed to the same weather conditions [96].
a) System Performance Analysis for HILP Scenarios Selection
Next, considering the aforementioned assumptions, when a severe wind event arises, the supply
availability and the corresponding system performance progresses through multiple phases along Ω𝑇 ,
leading to the typical progression curve for the system performance (SPF) during a severe event
depicted in Figure 5. 2 [94], [56].
Line(s) or branch(s) All lines recovered and
collapse and network is the operation return to
clustered into MGs for normal grid
resisting the HILP event connected mode
SPF
Before Disturbance Restoration start Normal
System Performance SPF(t)
event progress and progress operation

Lin e fail Lin es repair
SPF0 progress progress
MGs operated MGs operated at Update M Gs

dependent to worst deg raded formation based on
N-K contingencies stage line(s ) recovered
Restoration First line
start rep aired
SPFmin
Phase I Phase II
event
T T repair
t0 te td tr' tr Time
FIGURE 5. 2: Evolution of network’s system performance during a severe event
In this perspective, the system performance can be represented as a function of the electricity supply
level at time 𝑡, being divided into Phase I and II, where Phase I represents the disturbance progress
period between the event start, 𝑡 𝑒 , and maximum system degradation state, 𝑡 𝑑 , i.e. 𝑇 𝑒𝑣𝑒𝑛𝑡 = 𝑡 𝑒 − 𝑡 𝑑 ,
and Phase II depicts the recovery period between the maximum degradation time until reparations
conclusion, i.e., 𝑇 𝑟𝑒𝑝𝑎𝑖𝑟 = 𝑡 𝑑 − 𝑡 𝑟 . Here it is important to note that there is typically a lead time in
85
Phase II between the moment where the event disturbance progression stops and the time where
′
repairs’ impact is effectively observed, i.e., 𝑡 𝑟 . Detailed descriptions of Phase I and II leading to the
developed weather-related algorithm for HILP scenarios selection and component availability matrix
𝐀𝐕 derivation are following presented.
b) Phase I: Disturbance Period
Assuming the event start and progression, during Phase I the affected system clusters into multiple
MGs with topologies respecting local resources availability and isolate switches allocated in the
network. The formation of MGs allows the system to improve its withstanding ability against the
N-K contingencies cause by the severe wind and enhance the maintenance of the loads service. In
this sense, given this work focus on resilience-based planning, the single most vulnerable wind
scenario needs to be identified for each season in order to stress the system performance under worst-
case condition, i.e., 𝒗𝑤𝑜𝑟𝑠𝑡
𝑠𝑒 . For this, based on works [97], [98] wind gusts can be calculated using
(5.7), where the system most vulnerable scenario under a resilience perspective is given by the wind
𝐦
scenario within matrix 𝐯𝑠𝑒 containing the highest wind velocity.
𝑔𝑢𝑠𝑡
𝒗𝑠𝑒 = 𝑘 𝑔𝑢𝑠𝑡 𝒗𝑤𝑜𝑟𝑠𝑡
𝑠𝑒 ,∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.7)
Following, the failure probability of overhead lines for season 𝑠𝑒 can be extracted by mapping poles
𝑔𝑢𝑠𝑡
and conductors’ fragility curve for the calculated wind gust profile 𝒗𝑠𝑒 . The fragility curve can be
generated using the normal cumulative distribution function as generically illustrated in Figure 5. 3
for the distribution overhead line depicted in [95].
FIGURE 5. 3: Generic fragility curve for single pole and conductor against wind gust
Here, it is worth noting that a branch fails if a conductor, or any pole overhead line, connected between
two adjacent nodes fails. In this sense, the total failure probability of the distribution overhead line
should be calculated using (5.8)-(5.11).
𝑝𝑜𝑙𝑒 𝑔𝑢𝑠𝑡 𝑔𝑢𝑠𝑡 𝑝𝑜𝑙𝑒 𝑝𝑜𝑙𝑒

𝜋𝑖,𝑡,𝑠𝑒 (𝑣𝑡,𝑠𝑒 ) = 𝑐𝑑𝑓 𝑛𝑜𝑟 (𝑣𝑡,𝑠𝑒 , 𝜇𝑖 , 𝜎𝑖 ) ,∀𝑖 ∈ Ω𝐿𝐼𝑁 , ∀𝑡 ∈ Ω𝑇 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.8)
86
𝑐𝑜𝑛𝑑 𝑔𝑢𝑠𝑡 𝑔𝑢𝑠𝑡
𝜋𝑖,𝑡,𝑠𝑒 (𝑣𝑡,𝑠𝑒 ) = 𝑐𝑑𝑓 𝑛𝑜𝑟 (𝑣𝑡,𝑠𝑒 , 𝜇𝑖𝑐𝑜𝑛𝑑 , 𝜎𝑖𝑐𝑜𝑛𝑑 ) ,∀𝑖 ∈ Ω𝐿𝐼𝑁 , ∀𝑡 ∈ Ω𝑇 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.9)
𝑝𝑜𝑙𝑒
𝑁𝑖
𝑝𝑜𝑙𝑒𝑠 𝑔𝑢𝑠𝑡 𝑝𝑜𝑙𝑒 𝑔𝑢𝑠𝑡
𝜋𝑖,𝑡,𝑠𝑒 (𝑣𝑡,𝑠𝑒 ) = 1 − [(1 − 𝜋𝑖,𝑡,𝑠𝑒 (𝑣𝑡,𝑠𝑒 )) ] ,∀𝑖 ∈ Ω𝐿𝐼𝑁 , ∀𝑡 ∈ Ω𝑇 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.10)
𝑐𝑜𝑛𝑑 𝑔𝑢𝑠𝑡
𝜋𝑖,𝑡,𝑠𝑒 (𝑣𝑡,𝑠𝑒 )
𝑝𝑜𝑙𝑒𝑠 𝑔𝑢𝑠𝑡
𝑏𝑟 𝑔𝑢𝑠𝑡 +𝜋𝑖,𝑡,𝑠𝑒 (𝑣𝑡,𝑠𝑒 )
𝜋𝑖,𝑡,𝑠𝑒 (𝑣𝑡,𝑠𝑒 ) = 𝑐𝑜𝑛𝑑 𝑔𝑢𝑠𝑡 ,∀𝑖 ∈ Ω𝐿𝐼𝑁 , ∀𝑡 ∈ Ω𝑇 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.11)
𝜋𝑖,𝑡,𝑠𝑒 (𝑣𝑡,𝑠𝑒 )
−( 𝑝𝑜𝑙𝑒𝑠 𝑔𝑢𝑠𝑡 )
( × 𝜋𝑖,𝑡,𝑠𝑒 (𝑣𝑡,𝑠𝑒 ) )
𝑔𝑢𝑠𝑡
Eqs. (5.8) and (5.9) define the failure probability for a single pole and conductor during 𝑣𝑡,𝑠𝑒 based
on the normal cumulative distribution function with parameter 𝜇 𝑝𝑜𝑙𝑒 , 𝜎 𝑝𝑜𝑙𝑒 , 𝜇𝑐𝑜𝑛𝑑 and 𝜎 𝑐𝑜𝑛𝑑 . Eqs.
(5.10) and (5.11) determine the general formulation for dealing with the failure probability of branch
𝑝𝑜𝑙𝑒
𝑖 according to the quantity of poles 𝑁𝑖 for the overhead line of branch 𝑖, assuming that the poles
𝑝𝑜𝑙𝑒
and conductor in the distribution system have same characteristic and age. For this, 𝑁𝑖 can be
estimated based on the span length of the distribution overhead line [99]. For distribution systems,
the span length is typically between 100 ft to 150 ft, significantly shorter when compared to those
used for the transmission system [100].
The failure probability obtained for the branches is stored in matrix 𝛑𝑠𝑒 as shown in (5.12). Next,
similar to the work in [95], non-vulnerable branches are identified and disregarded by
𝑏𝑟
𝛑𝑠𝑒 = 0| 𝛑𝑠𝑒 (𝜋𝑖,𝑡,𝑠𝑒 < 𝜋 𝑏𝑟,𝑛𝑣𝑢𝑙 ). Following, for enlarging the possibility space of N-K
𝑝𝑟𝑒 I
contingencies, 𝑁𝑠 scenarios of 𝐀𝐕 are generated and represented by 𝐀𝐕𝑠,𝑠𝑒 considering Phase I.
𝑏𝑟 𝑏𝑟
𝜋𝑖,𝑡,𝑠𝑒 ⋯ 𝜋𝑖,𝑁 𝑇 ,𝑠𝑒
𝑔𝑢𝑠𝑡
𝛑𝑠𝑒 (𝒗𝑠𝑒 ) =[ ⋮ ⋱ ⋮ ] , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.12)
𝜋𝑛𝑏𝑟𝑏𝑟 ,𝑡,𝑠𝑒 ⋯ 𝜋𝑛𝑏𝑟𝑏𝑟 ,𝑁𝑇 ,𝑠𝑒
𝑏𝑟 𝑏𝑟
𝑎𝑖,𝑡,𝑠,𝑠𝑒 ⋯ 𝑎𝑖,𝑁 𝑇 ,𝑠,𝑠𝑒
I 𝑝𝑟𝑒
𝐀𝐕𝑠,𝑠𝑒 =[ ⋮ ⋱ ⋮ ] ,∀𝑖 ∈ Ω𝐿𝐼𝑁 , ∀𝑡 ∈ Ω𝑇 , ∀𝑠 ∈ Ω𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.13)
𝑏𝑟 𝑏𝑟
𝑎𝑛𝑏𝑟 ,𝑡,𝑠,𝑠𝑒 ⋯ 𝑎𝑛𝑏𝑟,𝑁𝑇 ,𝑠,𝑠𝑒
𝑏𝑟 𝑏𝑟
where, 𝑎𝑖,𝑡,𝑠,𝑠𝑒 is 0 if the 𝑖 𝑡ℎ branch fails at time 𝑡 for scenario 𝑠 of season 𝑠𝑒, otherwise, 𝑎𝑖,𝑡,𝑠,𝑠𝑒 = 1.
𝑏𝑟 𝑏𝑟
𝑎𝑖,𝑡,𝑠,𝑠𝑒 can be determined comparing the failure probability 𝜋𝑖,𝑡,𝑠,𝑠𝑒 to the generated uniformed
random binary number, as expressed by (5.14).
87
𝑏𝑟 𝑏𝑟
𝑏𝑟
𝑎𝑖,𝑡−1,𝑠,𝑠𝑒 × 1, 𝜋𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑟𝑖,𝑡,𝑠 𝑝𝑟𝑒
𝑎𝑖,𝑡,𝑠,𝑠𝑒 = { 𝑏𝑟 𝑏𝑟 ,∀𝑖 ∈ Ω𝐿𝐼𝑁 , ∀𝑡 ∈ Ω𝑇 , ∀𝑠 ∈ Ω𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.14)
𝑎𝑖,𝑡−1,𝑠,𝑠𝑒 × 0, 𝜋𝑖,𝑡,𝑠,𝑠𝑒 > 𝑟𝑖,𝑡,𝑠
c) Phase II: Restoration Period
Once the system reaches the maximum degradation condition, Phase II begins repairing the system
until all branches are fully recovered at 𝑡 𝑟 and the system is back to normal operation, i.e., grid
connected. In this sense, considering the time to repair (TTR) overhead lines dependency to the wind
velocity at the time of failure, i.e., 𝑡 = 𝑡𝑖,𝑠,𝑠𝑒 [94], while aware that due to safety reasons, there is
𝑑
no repair actions implemented until the end of the event at 𝑡𝑠,𝑠𝑒 [94]. The weather-related restoration
𝑔𝑢𝑠𝑡
time expressed by (5.15) and (5.16) is employed for determining TTR, and 𝜑𝑏𝑟 (𝑣𝑡,𝑠,𝑠𝑒 ) is determined
𝒃𝒓,𝒇𝒂𝒊𝒍
using uniform distributed sampling 𝜑~𝑈[ℎ1 , ℎ2 ] as depicted in (5.17) [94],[101]. Once 𝑻𝑻𝑹𝑠,𝑠𝑒
𝑑 𝑏𝑟,𝑟𝑒𝑝
and 𝑡𝑠,𝑠𝑒 are determined, the recovered time for a single branch 𝑖 can be determined by 𝑡𝑖,𝑠,𝑠𝑒 =
𝑑 𝑏𝑟,𝑓𝑎𝑖𝑙 𝑏𝑟,𝑟𝑒𝑝
𝑏𝑟 𝑏𝑟,𝑟𝑒𝑝
𝑏𝑟,𝑛𝑒𝑤
𝑡𝑠,𝑠𝑒 + 𝑇𝑇𝑅𝑖,𝑠,𝑠𝑒 . Thus, at 𝑡𝑖,𝑠,𝑠𝑒 , 𝑎𝑖,𝑡,𝑠,𝑠𝑒 (𝑡𝑖,𝑠,𝑠𝑒 ) = 𝑎𝑖,𝑡,𝑠,𝑠𝑒 = 1, which represents that branch 𝑖
is recovered and turned back to the online state. Finally, the branch availability matrix considering
𝐈−𝐈𝐈,𝐨𝐫𝐢𝐠 I 𝑏𝑟,𝑛𝑒𝑤
Phase I and II 𝐀𝐕𝑠,𝑠𝑒 can be obtained updating 𝐀𝐕𝑠,𝑠𝑒 with the new 𝑎𝑖,𝑡,𝑠,𝑠𝑒 information as shown
in (5.15)-(5.18).
𝒃𝒓,𝒇𝒂𝒊𝒍 𝑏𝑟,𝑓𝑎𝑖𝑙 𝑏𝑟,𝑓𝑎𝑖𝑙 𝑏𝑟,𝑓𝑎𝑖𝑙

𝑻𝑻𝑹𝑠,𝑠𝑒 = [𝑇𝑇𝑅1,𝑠,𝑠𝑒 , 𝑇𝑇𝑅2,𝑠,𝑠𝑒 , … , 𝑇𝑇𝑅𝑛𝑏𝑟,𝑠,𝑠𝑒 ] , ∀𝑠 ∈ Ω′𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.15)
𝑔𝑢𝑠𝑡
𝑏𝑟,𝑓𝑎𝑖𝑙 𝜑𝑏𝑟 (𝑣 𝑏𝑟,𝑓𝑎𝑖𝑙0 ) 𝑓𝑎𝑖𝑙 𝑝𝑟𝑒
𝑇𝑇𝑅𝑖,𝑠,𝑠𝑒 = 𝑟𝑜𝑢𝑛𝑑 ( 𝑡=𝑡𝑖,𝑠,𝑠𝑒 ,𝑠,𝑠𝑒 ) ,∀𝑖 ∈ Ω𝐿𝐼𝑁 , ∀𝑠 ∈ Ω𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.16)
𝑛𝑤
× 𝑀𝑇𝑇𝑅
𝑔𝑢𝑠𝑡 𝑙𝑜𝑤
1 , 𝑖𝑓 𝑣𝑡,𝑠,𝑠𝑒 < 𝑣𝑐𝑟𝑖
𝑔𝑢𝑠𝑡 𝑙𝑜𝑤 𝑔𝑢𝑠𝑡 ℎ𝑖𝑔ℎ
𝜑𝑏𝑟 (𝑣𝑡,𝑠,𝑠𝑒 ) = 𝜑~𝑈[ℎ1𝑙𝑜𝑤 , ℎ2𝑙𝑜𝑤 ], 𝑖𝑓 𝑣𝑐𝑟𝑖 ≤ 𝑣𝑡,𝑠,𝑠𝑒 ≤ 𝑣𝑐𝑟𝑖
ℎ𝑖𝑔ℎ ℎ𝑖𝑔ℎ 𝑔𝑢𝑠𝑡 ℎ𝑖𝑔ℎ
{𝜑~𝑈[ℎ1 , ℎ2 ], 𝑖𝑓 𝑣𝑡,𝑠,𝑠𝑒 > 𝑣𝑐𝑟𝑖
𝑓𝑎𝑖𝑙 𝑝𝑟𝑒
,∀𝑖 ∈ Ω𝐿𝐼𝑁 , ∀𝑠 ∈ Ω𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.17)
𝑏𝑟
𝑎1,1,𝑠,𝑠𝑒 , 𝑎𝑏𝑟,𝑛𝑒𝑤
, … , 𝑎1,𝑁𝑇 ,𝑠,𝑠𝑒
1,𝑡=𝑡𝑖,𝑠,𝑠𝑒 ,𝑠,𝑠𝑒
𝐈−𝐈𝐈,𝐨𝐫𝐢𝐠
𝑏𝑟
𝑎2,1,𝑠,𝑠𝑒 , 𝑎𝑏𝑟,𝑛𝑒𝑤
, … , 𝑎2,𝑁𝑇 ,𝑠,𝑠𝑒 ′′,𝑜𝑟𝑖𝑔
𝐀𝐕𝑠,𝑠𝑒 = 2,𝑡=𝑡𝑖,𝑠,𝑠𝑒 ,𝑠,𝑠𝑒 ,∀𝑠 ∈ Ω𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.18)
⋮
𝑎𝑛𝑏𝑟𝑏𝑟 ,1,𝑠,𝑠𝑒 , 𝑎𝑏𝑟,𝑛𝑒𝑤 𝑏𝑟,𝑟𝑒𝑝 , … , 𝑎𝑛𝑏𝑟,𝑛𝑒𝑤
𝑏𝑟 ,𝑁𝑇 ,𝑠,𝑠𝑒 ]
[ 𝑛𝑏𝑟 ,𝑡=𝑡𝑖,𝑠,𝑠𝑒 ,𝑠,𝑠𝑒
88
Next, the highest impact scenarios are determined for each season se until reaching the desired
𝐻𝐼
number of samples, i.e., 𝑁 − (𝐾𝑠,𝑠𝑒 ) are chosen first, where 𝑁 − (𝐾𝑠,𝑠𝑒 ) denotes N-K contingencies
for scenarios 𝑠, 𝑠𝑒, 𝐾𝑠,𝑠𝑒 corresponds to the number of fault locations occurred in the network for
scenarios 𝑠, 𝑠𝑒. In case the highest impact selected scenarios do not reach the expected number of
𝐻𝐼 𝐻𝐼
samples, available space(s) are fulfilled by the scenarios 𝑁 − (𝐾𝑠,𝑠𝑒 − 1), then 𝑁 − (𝐾𝑠,𝑠𝑒 − 2) until
the N-1 contingency.
𝐈−𝐈𝐈 𝐈−𝐈𝐈,𝐨𝐫𝐢𝐠
𝐀𝐕𝑠,𝑠𝑒 = 𝐀𝐕𝑠,𝑠𝑒 {𝑁 − (𝐾)} ,∀𝑠 ∈ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.19)
The procedure for generating the 𝐀𝐕 matrix is illustrated in Figure 5. 4.
START
Import the wind scenarios vmse
Select the most vulnerable

series of windspeed vworstse
Calculate the wind gust vgustse and

extract the matrix of failure probability
πse referring to the calculated wind gust
Phase I
Generate a set of uniform random (td - te)
numbers ri,t,s
For t belonging to T compare ri,t,s

to πse for determining AVIs,se
Store AVIs,se in th 3D matrix

AVI for season se
s=s+1 s>Ns
yes
Identify t d and generate TTRbr,fail Phase II

for scenario s season se, forming (te - tr)
AVI-II,origs,se
se=se+1 se>Nse
yes
Obtain AVI-II,orig and perform the

selection process for AVI-II
END
FIGURE 5. 4: Weather-related algorithm for generating branch availability matrix
89
5.4 Stage II: Planning Problem Formulation
In this section the problem formulation is detailed for both outer- and lower-levels, including
respective discussions on objective functions, constraints, and decision variables.
5.4.1 Outer Level: Optimal Allocation Level
The outer level problem formulation is responsible for the allocation of the planning elements include
the DDGs, WTs, PVs, ESSs and ISWs. For this, the following multi-objective problem formulations
is developed.
5.4.1.1 Objective Function for Outer Level
The multi-objective function for the outer level considers three main perspectives, i.e., 1) resilience
enhancement, 2) economic benefit, and 3) environmental impact mitigation. The proposed
optimization problem is solved using a multi-objective optimization. This allows the solution to this
problem to entail the best trade-off between these possibly conflicting objectives and enable to select
the optimal design under each objective’s consideration i.e., thought the pareto set. In this work, each
objective detailed described in the following subsections.
a) Resilience Index
The resilience index is responsible for ensuring the system capacity to withstand HILP events. For
this, the system’s loss performance considering the duration of degradation and the importance of
rapid recovery has been used [56]. The resilience index considering Phase I and II can be expressed
using (5.20).
1 𝑡 𝑟 𝑆𝑃𝐹0 (𝑡)−𝑆𝑃𝐹(𝑡)
𝑅𝐸𝑆𝐼 = ∫𝑡 𝑒 [ ] 𝑑𝑡 (5.20)
𝑡 𝑟 −𝑡 𝑒 𝑆𝑃𝐹(𝑡)
Then, by considering the set of scenarios 𝑠 and season 𝑠𝑒 according to Figure 5. 1, eq. (5.20) can be
re-formulated into (5.21)-(5.22). This perspective allows for combining the multiple inner level
season and scenarios information, i.e., ∀𝑠 ∈ Ω′′
𝑠 and ∀𝑠𝑒 ∈ Ω𝑠𝑒 , in a way to determine a single value
for the outer level resilience index.
𝐿
∑𝑖∈Ω𝐵𝑈𝑆 𝑃𝑖,𝑡,𝑠,𝑠𝑒
( )
1 𝑡𝑟 − ∑𝑖∈Ω𝐵𝑈𝑆 (1−𝑏𝑐𝑢𝑟𝑡 𝐿
𝑖,𝑡,𝑠,𝑠𝑒 )𝑃𝑖,𝑡,𝑠,𝑠𝑒
𝑅𝐸𝑆𝐼𝑠,𝑠𝑒 = 𝑟 −𝑡 𝑒 ∫𝑡 𝑒𝑠,𝑠𝑒 𝐿
∑𝑖∈Ω𝐵𝑈𝑆 𝑃𝑖,𝑡,𝑠,𝑠𝑒
𝑑𝑡 ,∀𝑠 ∈ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.21)
𝑡𝑠,𝑠𝑒 𝑠,𝑠𝑒 𝑠,𝑠𝑒
( )
90
1
𝑅𝐸𝑆𝐼 𝑂𝐿𝐸𝑉 = ∑𝑠∈Ω𝑠𝑒 ∑ ′′ 𝑅𝐸𝑆𝐼𝑠,𝑠𝑒 (5.22)
𝑁𝑠′′ 𝑠∈Ω𝑠
b) Economic Index
The second objective function relates to the economic benefit, i.e., capital cost, installation cost, and
O&M cost for ISWs [91] and DERs [83], [85], [102]. Where, eq. (5.23) defines the total cost as the
second objective function for the outer level. Eqs. (5.24)-(5.26) determine the capital and O&M cost
for the DDG, while (5.27) normalizes the cost (5.24) and (5.26) for DDG units by considering the
planning lifetime 𝑇𝐿. Similarly, eqs. (5.28)-(5.33) summarize the capital and O&M cost for the
allocated WTs and PVs. In term of the ESS, (5.34) and (5.35) determine the capital cost depending
to ESS power and energy density, respectively. The annual O&M cost and the installation cost for
the ESS can be calculated using (5.36) and (5.37), respectively. In (5.38) the normalized cost for ESS
considering 𝑇𝐿 is expressed. Finally, the cost for ISWs can be determined using (5.39)-(5.41).
𝑇𝐶 𝑂𝐿𝐸𝑉 = (𝐶 𝐷𝐷𝐺 + 𝐶 𝑊𝑇 + 𝐶 𝑃𝑉 + 𝐶 𝐸𝑆𝑆 + 𝐶 𝐼𝑆𝑊 ) (5.23)
𝐶𝑃𝐶 𝐷𝐷𝐺 = ∑∀𝑖∈Ω𝑐𝑎𝑛

𝐷𝐷𝐺
(𝑃𝑖𝐷𝐷𝐺,𝑠𝑖𝑧𝑒 𝜎𝐼𝑁𝑉
𝐷𝐷𝐺 ) (5.24)
𝐷𝐷𝐺 𝜎 𝑓𝑢𝑒𝑙 ⁄𝜂 𝐷𝐷𝐺 𝐷𝐷𝐺

𝑂𝑀𝐶𝑠,𝑠𝑒 = 𝑆𝑏𝑎𝑠𝑒 ( 𝐷𝐷𝐺 ) ∑∀𝑖∈Ω𝐷𝐷𝐺 ∑∀𝑡∈Ω𝑇 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ,∀𝑠 ∈ Ω′𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.25)
+𝜎𝑂&𝑀
365
𝐴𝑂𝑀𝐶 𝐷𝐷𝐺 = 4
𝑤𝑖𝑠𝑜
∑∀𝑠𝑒∈ Ω𝑠𝑒 ∑∀𝑠 ∈ Ω′𝑠(𝜋𝑠,𝑠𝑒 𝐷𝐷𝐺
𝑂𝑀𝐶𝑠,𝑠𝑒 ) (5.26)
𝐶 𝐷𝐷𝐺 = 𝐶𝑃𝐶 𝐷𝐷𝐺 + 𝑇𝐿 × 𝐴𝑂𝑀𝐶 𝐷𝐷𝐺 (5.27)
𝐶𝑃𝐶 𝑃𝑉 = ∑∀𝑖∈Ω𝑐𝑎𝑛
𝑃𝑉
(𝑃𝑖𝑃𝑉,𝑠𝑖𝑧𝑒 𝜎𝐼𝑁𝑉
𝑃𝑉 ) (5.28)
𝐴𝑂𝑀𝐶 𝑃𝑉 = ∑∀𝑖∈Ω𝑐𝑎𝑛
𝑃𝑉
(𝑃𝑖𝑃𝑉,𝑠𝑖𝑧𝑒 𝜎𝑂&𝑀
𝑃𝑉
) (5.29)
𝐶 𝑃𝑉 = 𝐶𝑃𝐶 𝑃𝑉 + 𝑇𝐿 × 𝐴𝑂𝑀𝐶 𝑃𝑉 (5.30)
𝐶𝑃𝐶 𝑊𝑇 = ∑∀𝑖∈Ω𝑐𝑎𝑛
𝑊𝑇
(𝑃𝑖𝑊𝑇,𝑠𝑖𝑧𝑒 𝜎𝐼𝑁𝑉
𝑊𝑇 ) (5.31)
𝐴𝑂𝑀𝐶 𝑊𝑇 = ∑∀𝑖∈Ω𝑐𝑎𝑛
𝑊𝑇
(𝑃𝑖𝑊𝑇,𝑠𝑖𝑧𝑒 𝜎𝑂&𝑀
𝑊𝑇
) (5.32)
𝐶 𝑊𝑇 = 𝐶𝑃𝐶 𝑊𝑇 + 𝑇𝐿 × 𝐴𝑂𝑀𝐶 𝑊𝑇 (5.33)
91
𝐶𝑃𝐶𝐸𝑆𝑆,𝑃 = ∑∀𝑖∈Ω𝑐𝑎𝑛
𝐸𝑆𝑆
(𝑃𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 𝜎𝐼𝑁𝑉
𝐸𝑆𝑆,𝑃
) (5.34)
𝐶𝑃𝐶𝐸𝑆𝑆,𝐸 = ∑∀𝑖∈Ω𝑐𝑎𝑛
𝐸𝑆𝑆
(𝐸𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 𝜎𝐼𝑁𝑉
𝐸𝑆𝑆,𝐸
) (5.35)
𝐴𝑂𝑀𝐶 𝐸𝑆𝑆 = ∑∀𝑖∈Ω𝑐𝑎𝑛

𝐸𝑆𝑆
(𝑃𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 𝜎𝑂&𝑀
𝐸𝑆𝑆 ) (5.36)
𝐼𝑁𝑆𝐶 𝐸𝑆𝑆 = ∑∀𝑖∈Ω𝑐𝑎𝑛

𝐸𝑆𝑆
(𝐸𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 𝜎𝐼𝑁𝑆
𝐸𝑆𝑆 ) (5.37)
𝐶 𝐸𝑆𝑆 = (𝐶𝑃𝐶𝐸𝑆𝑆,𝑃 + 𝑇𝐿 × 𝐴𝑂𝑀𝐶 𝐸𝑆𝑆 + 𝐶𝑃𝐶𝐸𝑆𝑆,𝐸 + 𝐼𝑁𝑆𝐶 𝐸𝑆𝑆 ) (5.38)
𝐶𝑃𝐶 𝐼𝑆𝑊 = ∑∀𝑖∈Ω𝑐𝑎𝑛

𝐼𝑆𝑊
(𝑏𝑖𝐼𝑆𝑊 𝜎𝐼𝑁𝑉
𝐼𝑆𝑊
) (5.39)
𝐴𝑂𝑀𝐶 𝐼𝑆𝑊 = 8760 ∑∀𝑖∈Ω𝑐𝑎𝑛

𝐼𝑆𝑊
(𝑏𝑖𝐼𝑆𝑊 𝜎𝑂&𝑀
𝐼𝑆𝑊
) (5.40)
𝐶 𝐼𝑆𝑊 = 𝐶𝑃𝐶 𝐼𝑆𝑊 + 𝑇𝐿 × 𝐴𝑂𝑀𝐶 𝐼𝑆𝑊 (5.41)
c) Environmental Impact Index
The environment impact index aims to reduce the carbon footprint due to energy purchases from the
main grid and energy generated by DDG [102], [103]. The annual carbon emission can be calculated
using the expressions below. Eqs. (5.42) and (5.43) are the CO2 emission quantity and their annualized
kg/year from the DDGs generation. Eqs. (5.44) and (5.45) represent the CO2 emission depending on
the power imported from the main grid. Finally, (5.46) defines the total annualized Carbon emission
as the third objective function for the outer level.
𝐷𝐷𝐺 𝐷𝐷𝐺 𝐷𝐷𝐺

𝐸𝑀𝐼𝑆𝑠,𝑠𝑒 = 𝐾𝐶𝑂2 𝑆𝑏𝑎𝑠𝑒 ∑∀𝑖∈Ω𝐷𝐷𝐺 ∑∀𝑡 ∈ Ω𝑇 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ,∀𝑠 ∈ Ω′𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.42)
365
𝐴𝐸𝑀𝐼𝑆 𝐷𝐷𝐺 = 4
𝑤𝑖𝑠𝑜
∑∀𝑠𝑒∈Ω𝑠𝑒 ∑∀𝑠∈Ω′𝑠(𝜋𝑠,𝑠𝑒 𝐷𝐷𝐺
𝐸𝑀𝐼𝑆𝑠,𝑠𝑒 ) (5.43)
𝐺𝑅𝐼𝐷 𝐺𝑅𝐼𝐷 𝐺𝑅𝐼𝐷+

𝐸𝑀𝐼𝑆𝑠,𝑠𝑒 = 𝐾𝐶𝑂2 𝑆𝑏𝑎𝑠𝑒 ∑∀𝑡∈Ω𝑇 𝑃𝑡,𝑠,𝑠𝑒 ,∀𝑠 ∈ Ω′𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.44)
365
𝐴𝐸𝑀𝐼𝑆 𝐺𝑅𝐼𝐷 = 4
𝑤𝑖𝑠𝑜
∑∀𝑠𝑒∈Ω𝑠𝑒 ∑∀𝑠∈Ω′𝑠(𝜋𝑠,𝑠𝑒 𝐺𝑅𝐼𝐷
𝐸𝑀𝐼𝑆𝑠,𝑠𝑒 ) (5.45)
𝐴𝐸𝑀𝐼𝑆 𝑂𝐿𝐸𝑉 = 𝐴𝐸𝑀𝐼𝑆 𝐷𝐷𝐺 + 𝐴𝐸𝑀𝐼𝑆 𝐺𝑅𝐼𝐷 (5.46)
92
5.4.1.2 Constraints for Outer Level
Next, system’s constraints and decision variables for optimizing the outer level are following
depicted.
a) Discrete Size for DERs Constraints
The DERs size were discretized with a fixed step [103].
𝑃𝑖𝐷𝐷𝐺,𝑠𝑖𝑧𝑒 = 𝑏𝑖𝐷𝐷𝐺,𝑠𝑖𝑧𝑒 𝑃𝑠𝑡𝑒𝑝

𝐷𝐷𝐺
,∀𝑖 ∈ Ω𝑐𝑎𝑛
𝐷𝐷𝐺 (5.47)
𝑃𝑖𝑊𝑇,𝑠𝑖𝑧𝑒 = 𝑏𝑖𝑊𝑇,𝑠𝑖𝑧𝑒 𝑃𝑠𝑡𝑒𝑝

𝑊𝑇
𝑊𝑇 (5.48)
𝑃𝑖𝑃𝑉,𝑠𝑖𝑧𝑒 = 𝑏𝑖𝑃𝑉,𝑠𝑖𝑧𝑒 𝑃𝑠𝑡𝑒𝑝

𝑃𝑉
𝑃𝑉 (5.49)
𝑃𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 = 𝑏𝑖𝐸𝑆𝑃,𝑠𝑖𝑧𝑒 𝑃𝑠𝑡𝑒𝑝

𝐸𝑆𝑆
𝐸𝑆𝑆 (5.50)
𝐸𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 = 𝑟𝑖𝐸𝑆𝑆 × 𝑃𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 ,∀𝑖 ∈ Ω𝑐𝑎𝑛

𝐸𝑆𝑆 (5.51)
b) Maximum Bus Power Allowable Constraints
The DER(s) connected to an individual bus must respected the technical constraints defined by (5.52)
[103]:
∑∀𝐷𝐸𝑅∈{𝐷𝐷𝐺,𝑃𝑉,𝑊𝑇,𝐸𝑆𝑆} 𝑃𝑖𝐷𝐸𝑅,𝑠𝑖𝑧𝑒 ≤ 𝑃𝑖𝑏𝑢𝑠,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 (5.52)
c) Installation Discrete Binary Constraints
0 ≤ 𝑏𝑖𝐷𝐷𝐺,𝑠𝑖𝑧𝑒 ≤ 𝑃𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 /𝑃𝑠𝑡𝑒𝑝

𝐷𝐷𝐺
𝐷𝐷𝐺 (5.53)
0 ≤ 𝑏𝑖𝑊𝑇,𝑠𝑖𝑧𝑒 ≤ 𝑃𝑖𝑊𝑇,𝑚𝑎𝑥 /𝑃𝑠𝑡𝑒𝑝

𝑊𝑇
𝑊𝑇 (5.54)
0 ≤ 𝑏𝑖𝑃𝑉,𝑠𝑖𝑧𝑒 ≤ 𝑃𝑖𝑃𝑉,𝑚𝑎𝑥 /𝑃𝑠𝑡𝑒𝑝

𝑃𝑉
𝑃𝑉 (5.55)
0 ≤ 𝑏𝑖𝐸𝑆𝑃,𝑠𝑖𝑧𝑒 ≤ 𝑃𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 /𝑃𝑠𝑡𝑒𝑝

𝐸𝑆𝑆
𝐸𝑆𝑆 (5.56)
0 ≤ 𝑏𝑖𝐼𝑆𝑊 ≤ 1 ,∀𝑖 ∈ Ω𝑐𝑎𝑛

𝐼𝑆𝑊 (5.57)
93
5.4.1.3 Decision Variables for Outer Level
The decision variables for optimizing the outer level are summarized in Table 5. 1. Here one should
note that variables associated with each planning element are respectively set to zero for the case that
the optimization decides to not place these elements on the candidate buses or branches. Otherwise,
these variables are the integer numbers respecting to the upper bounds defined in (53)-(57), and the
binary value ‘1’ for 𝑏𝑖𝐼𝑆𝑊 .
TABLE 5. 1: Decision Variables Type and Set Belonging for the Outer Level Problem
Variables type Set
𝑿𝑶𝑳𝑬𝑽 ∀𝑖 ∈
𝑏𝑖𝐷𝐷𝐺,𝑠𝑖𝑧𝑒 ℤ Ω𝑐𝑎𝑛
𝐷𝐷𝐺
𝑏𝑖𝑊𝑇,𝑠𝑖𝑧𝑒 ℤ Ω𝑐𝑎𝑛
𝑊𝑇
𝑏𝑖𝑃𝑉,𝑠𝑖𝑧𝑒 ℤ Ω𝑐𝑎𝑛
𝑃𝑉
𝑏𝑖𝐸𝑆𝑃,𝑠𝑖𝑧𝑒 ℤ Ω𝑐𝑎𝑛
𝐸𝑆𝑆
𝑏𝑖𝐼𝑆𝑊 ℤ Ω𝑐𝑎𝑛
𝐼𝑆𝑊
5.4.2 Inner Level: Optimal Power Flow for Microgrid
This level presents the problem to be solved using the proposed machine learning based technique.
Still, first a pre-problem formulated as a MILP model similar to the work in [3] is developed. This
inner level optimization solutions provide the samples solutions needed to train the proposed machine
learning-based strategy using the upper-level solutions as inputs. For this, this problem models the
ability of microgrids to operate in both grid-connected and IS mode, depending on the operation status
of the main isolate switch (ISW sub) as shown in Figure 5. 5. Details on the problem formulation and
optimization strategy including objective function, constraints and decision variables are following
depicted.
MICROGRID
kW kW
MW kVAR
MVAR
ISWsub kW
BUSBAR PCC MG
SUBSTATION kVAR kW
Candidate branch
Candidate bus
Microgrid PCC
DDG Substation
FIGURE 5. 5: Basic scheme of the single microgrid base 𝐈𝐒𝐖𝐬𝐮𝐛
94
5.4.2.1 Objective Function and Constraints for Inner Level
The objective function formulated for the optimal operation in the inner level are expressed by (5.58)-
(5.59). First, in (5.58) the objective function is obtained normalizing the four terms as a single total
operation cost to be minimized. The first term in (5.58) represents the cost of power exchange between
the microgrid and the main grid, which can be determined using (5.59). Then, the second term is
given by (5.60), which refers to the fuel consumption cost for the DDG units in use at the MGs [104],
[102]. As one may observe, (5.58), (5.59) and (5.60) are restricted to the grid-connected mode due to
𝑚 𝑚
the association of the binary variable 𝑏𝑡,𝑠,𝑠𝑒 , i.e., 𝑏𝑡,𝑠,𝑠𝑒 is the binary number that denotes the operation
𝑚
status for ISW sub along Ω 𝑇 , where 𝑏𝑡,𝑠,𝑠𝑒 = 1 represents the microgrid operation in islanding mode,
𝑚 𝑚
otherwise 𝑏𝑡,𝑠,𝑠𝑒 = 0 denotes the system operation in grid-connected mode. In this sense, 𝑏𝑡,𝑠,𝑠𝑒 can
be used as multiplication and integrated in the MILP model for allowing the objective function and
constraints pertaining to the operation mode for microgrid. The third term given by (5.61) represents
the cost for load curtailment, being restricted for islanded mode only [26]. The last term depicted in
(5.62) represents the voltage violation index. This term seeks to improve the voltage quality of the
microgrid during both operation modes. For this, the following assumptions are made: 1) during grid
connected mode, the DDG units are operated in PQmode [52], [54], 2) during IS mode the DDG units
are operated under a secondary control layer with 𝑉 ∗∗ identically set for all DDG units.
𝑀𝐼𝐿𝑃 𝑃𝐶𝐶 𝑚 𝐷𝐷𝐺 𝑚 𝐶𝑈𝑅𝑇 𝑚 𝑉𝐼

𝑂𝐹𝑠,𝑠𝑒 = ∑𝑡∈𝛺𝑇 [𝐶𝑡,𝑠,𝑠𝑒 (1 − 𝑏𝑡,𝑠,𝑠𝑒 ) + 𝐶𝑡,𝑠,𝑠𝑒 (1 − 𝑏𝑡,𝑠,𝑠𝑒 ) + 𝐶𝑡,𝑠,𝑠𝑒 𝑏𝑡,𝑠,𝑠𝑒 + 𝐶𝑡,𝑠,𝑠𝑒 ]
,∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.58)
𝑃𝐶𝐶 𝑃𝐶𝐶 𝑃𝐶𝐶

𝐶𝑡,𝑠,𝑠𝑒 = 𝑆𝑏𝑎𝑠𝑒 (𝜎𝑡,𝑠,𝑠𝑒 𝑃𝑖=1,𝑡,𝑠,𝑠𝑒 )∆𝑡 ,∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.59)
𝐷𝐷𝐺 𝐷𝐷𝐺 𝐷𝐷𝐺

𝐶𝑡,𝑠,𝑠𝑒 = 𝑆𝑏𝑎𝑠𝑒 (∑𝑖∈Ω𝐷𝐷𝐺 𝑃𝑖,𝑡,𝑠,𝑠𝑒 (𝜎 𝑓𝑢𝑒𝑙 ⁄𝜂 𝐷𝐷𝐺 + 𝜎𝑂&𝑀 )) ∆𝑡
,∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.60)
𝐶𝑈𝑅𝑇
𝐶𝑡,𝑠,𝑠𝑒 = 𝑆𝑏𝑎𝑠𝑒 (𝜎 𝑐𝑢𝑟𝑡 ∑𝑖∈Ω𝐵𝑈𝑆 𝑏𝑐𝑢𝑟𝑡 𝑙𝑜𝑎𝑑 𝐿
𝑖,𝑡,𝑠,𝑠𝑒 𝑤𝑖 𝑃𝑖,𝑡,𝑠,𝑠𝑒 )∆𝑡
,∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.61)
𝑉𝐼
𝐶𝑡,𝑠,𝑠𝑒 𝑚
= 𝑉𝑏𝑎𝑠𝑒 (𝑃𝐶 𝑉𝐼 [∑𝑖∈𝛺𝐵𝑈𝑆((1 − 𝑏𝑡,𝑠,𝑠𝑒 𝑚
)𝑉 𝐺𝐶,𝑠𝑝𝑒𝑐 + 𝑏𝑡,𝑠,𝑠𝑒 𝑉𝑖∗∗ − 𝑉𝑖,𝑡,𝑠,𝑠𝑒 )])∆𝑡
,∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.62)
Subject to:
𝑉 𝑚𝑖𝑛 ≤ 𝑉𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑉 𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.63)
95
0 ≤ 𝑏𝑐𝑢𝑟𝑡 𝑚
𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑏𝑡,𝑠,𝑠𝑒 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.64)
𝑃𝑖,𝑡,𝑠,𝑠𝑒 + 𝑃𝑖,𝑡,𝑠,𝑠𝑒
𝑊𝑇,𝑢𝑠𝑒 𝑃𝑉,𝑢𝑠𝑒
∑𝑗∈Ω𝐵𝑈𝑆 𝐺𝑖𝑗 𝑉𝑖,𝑡,𝑠,𝑠𝑒 +𝑃𝑖,𝑡,𝑠,𝑠𝑒 + 𝑃𝑖,𝑡,𝑠,𝑠𝑒
𝑃𝑖,𝑡,𝑠,𝑠𝑒 = ( )=
− ∑𝑗∈Ω𝐵𝑈𝑆 𝐵𝑖𝑗 𝛿𝑗,𝑡,𝑠,𝑠𝑒 𝑑𝑐ℎ
+𝑃𝑖,𝑡,𝑠,𝑠𝑒 − 𝜂𝑖𝐸𝑆𝑆 𝑃𝑖,𝑡,𝑠,𝑠𝑒
𝑐ℎ
𝐿 𝑐𝑢𝑟𝑡
( −𝑃𝑖,𝑡,𝑠,𝑠𝑒 (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒 ) )
,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.65)
+𝑄𝑖,𝑡,𝑠,𝑠𝑒 + 𝑄𝑖,𝑡,𝑠,𝑠𝑒
− ∑𝑗∈Ω𝐵𝑈𝑆 𝐵𝑖𝑗 𝑉𝑖,𝑡,𝑠,𝑠𝑒 𝐸𝑆𝑆
𝑄𝑖,𝑡,𝑠,𝑠𝑒 =( )=( +𝑄𝑖,𝑡,𝑠,𝑠𝑒 )
− ∑𝑗∈Ω𝐵𝑈𝑆 𝐺𝑖𝑗 𝛿𝑗,𝑡,𝑠,𝑠𝑒
−𝑄𝑖,𝑡,𝑠,𝑠𝑒 (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒 )
,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.66)
𝑏𝑟
𝑃𝑘,𝑡,𝑠,𝑠𝑒 = (𝑔𝑖(𝑘)𝑗(𝑘) (𝑉𝑖,𝑡,𝑠,𝑠𝑒 − 𝑉𝑗,𝑡,𝑠,𝑠𝑒 ) + 𝑏𝑖(𝑘)𝑗(𝑘) (𝛿𝑖,𝑡,𝑠,𝑠𝑒 − 𝛿𝑗,𝑡,𝑠,𝑠𝑒 ))
,∀𝑘 ∈ Ω𝐿𝐼𝑁 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.67)
𝑏𝑟
𝑄𝑘,𝑡,𝑠,𝑠𝑒 = (𝑏𝑖(𝑘)𝑗(𝑘) (𝑉𝑖,𝑡,𝑠,𝑠𝑒 − 𝑉𝑗,𝑡,𝑠,𝑠𝑒 ) + 𝑔𝑖(𝑘)𝑗(𝑘) (𝛿𝑖,𝑡,𝑠,𝑠𝑒 − 𝛿𝑗,𝑡,𝑠,𝑠𝑒 ))
,∀𝑘 ∈ Ω𝐿𝐼𝑁 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.68)
𝑏𝑟
𝛼𝑛1 𝑃𝑘,𝑡,𝑠,𝑠𝑒 𝑏𝑟
+ 𝛽𝑛1 𝑄𝑘,𝑡,𝑠,𝑠𝑒 − 𝛾 1 𝐼𝑘𝑏𝑟,𝑚𝑎𝑥 𝑉𝑖(𝑘),𝑡,𝑠,𝑠𝑒 ≤ 0
𝑏𝑟
−𝛼𝑛1 𝑃𝑘,𝑡,𝑠,𝑠𝑒 𝑏𝑟
+ 𝛽𝑛1 𝑄𝑘,𝑡,𝑠,𝑠𝑒 − 𝛾1 𝐼𝑘𝑏𝑟,𝑚𝑎𝑥 𝑉𝑖(𝑘),𝑡,𝑠,𝑠𝑒 ≤ 0
𝐿𝑏𝑟
𝑘,𝑡,𝑠,𝑠𝑒 = 𝑡ℎ 𝑡ℎ 𝑡ℎ
𝑏𝑟
−𝛼𝑛4 𝑃𝑘,𝑡,𝑠,𝑠𝑒 𝑏𝑟
− 𝛽𝑛4 𝑄𝑘,𝑡,𝑠,𝑠𝑒 − 𝛾 4 𝐼𝑘𝑏𝑟,𝑚𝑎𝑥 𝑉𝑖(𝑘),𝑡,𝑠,𝑠𝑒 ≤ 0
𝑡ℎ𝑏𝑟 𝑡ℎ𝑏𝑟 𝑡ℎ
𝑏𝑟,𝑚𝑎𝑥
{ 𝛼𝑛4 𝑃𝑘,𝑡,𝑠,𝑠𝑒 − 𝛽𝑛4 𝑄𝑘,𝑡,𝑠,𝑠𝑒 − 𝛾 4 𝐼𝑘 𝑉𝑖(𝑘),𝑡,𝑠,𝑠𝑒 ≤ 0
,𝑛 = 1,2, … , 𝑛𝑝𝑤 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑘 ∈ Ω𝐿𝐼𝑁 , ∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.69)
𝐷𝐷𝐺 𝑀𝐺 𝑚 𝑝 𝑚 𝐷𝐷𝐺
𝑃𝑖,𝑡,𝑠,𝑠𝑒 = ([− 𝑘 𝑖𝜔 (𝛿𝑡,𝑠,𝑠𝑒 − 𝜋⁄2)⁄𝑚𝑖 ]𝑏𝑡,𝑠,𝑠𝑒 + (1 − 𝑏𝑡,𝑠,𝑠𝑒 )𝑃𝑖,𝑡,𝑠,𝑠𝑒 )
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.70)
𝑞
𝐷𝐷𝐺
𝑄𝑖,𝑡,𝑠,𝑠𝑒 = ([(𝑉𝑖∗ − 𝑉𝑖,𝑡,𝑠,𝑠𝑒 + 𝑢𝑖,𝑡,𝑠,𝑠𝑒
𝑣 𝑚
)⁄𝑛𝑖 ]𝑏𝑡,𝑠,𝑠𝑒 𝑚
+ (1 − 𝑏𝑡,𝑠,𝑠𝑒 𝐷𝐷𝐺
)𝑄𝑖,𝑡,𝑠,𝑠𝑒 )
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.71)
𝑚
𝑉𝑖,𝑡,𝑠,𝑠𝑒 𝑏𝑡,𝑠,𝑠𝑒 = 𝑉𝑖∗∗ 𝑏𝑡,𝑠,𝑠𝑒
𝑚
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.72)
𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 = 𝑃𝑖𝐷𝐷𝐺,𝑠𝑖𝑧𝑒 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 (5.73)
96
𝑝
𝑚𝑖 = ∆𝜔⁄𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 (5.74)
𝑞
𝑛𝑖 = ∆𝑉⁄2𝜒𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 (5.75)
𝐷𝐷𝐺
0 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 ,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.76)
𝜒𝑖𝐷𝐷𝐺,𝑚𝑖𝑛 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 ≤ 𝑄𝑖,𝑡,𝑠,𝑠𝑒

𝐷𝐷𝐺
≤ 𝜒𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.77)
𝐷𝐷𝐺
−𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝑡𝑔(cos−1 𝑃𝐹𝑖𝐷𝐷𝐺,4𝑡ℎ ) ≤ 𝑄𝑖,𝑡,𝑠,𝑠𝑒
𝐷𝐷𝐺 𝐷𝐷𝐺
≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝑡𝑔(cos −1 𝑃𝐹𝑖𝐷𝐷𝐺,1𝑠𝑡 )
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.78)
𝐷𝐷𝐺
𝛼𝑛1 𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝐷𝐷𝐺
+ 𝛽𝑛1 𝑄𝑖,𝑡,𝑠,𝑠𝑒 − 𝛾1 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 ≤ 0
𝐿𝐷𝐷𝐺
𝑖,𝑡,𝑠,𝑠𝑒 ={ 𝑡ℎ 𝑡ℎ 𝑡ℎ
𝐷𝐷𝐺
𝛼𝑛4 𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝐷𝐷𝐺
− 𝛽𝑛4 𝑄𝑖,𝑡,𝑠,𝑠𝑒 − 𝛾 4 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 ≤ 0
,𝑛 = 1,2, … , 𝑛𝑝𝑤 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.79)
𝑊𝑇,𝑢𝑠𝑒 𝑊𝑇
0 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ,∀𝑖 ∈ Ω𝑊𝑇 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.80)
𝑃𝑉,𝑢𝑠𝑒 𝑃𝑉
0 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ,∀𝑖 ∈ Ω𝑃𝑉 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.81)
𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 = 𝑃𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 ,∀𝑖 ∈ Ω𝐸𝑆𝑆 (5.82)
𝑑𝑐ℎ
0 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 𝑏𝐸𝑆𝑆
𝑖,𝑡,𝑠,𝑠𝑒 ,∀𝑖 ∈ Ω𝐸𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.83)
𝑐ℎ
0 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 (1 − 𝑏𝐸𝑆𝑆
𝑖,𝑡,𝑠,𝑠𝑒 )
,∀𝑖 ∈ Ω𝐸𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.84)
𝜒𝑖𝐸𝑆𝑆,𝑚𝑖𝑛 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 ≤ 𝑄𝑖,𝑡,𝑠,𝑠𝑒

𝐸𝑆𝑆
≤ 𝜒𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥
,∀𝑖 ∈ Ω𝐸𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.85)
𝑑𝑐ℎ
𝛼𝑛1 (𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝑐ℎ
− 𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝐸𝑆𝑆
) + 𝛽𝑛1 𝑄𝑖,𝑡,𝑠,𝑠𝑒 − 𝛾1 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 ≤ 0
𝑑𝑐ℎ
−𝛼𝑛1 (𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝑐ℎ
) + 𝛽𝑛1 𝑄𝑖,𝑡,𝑠,𝑠𝑒 − 𝛾 1 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 ≤ 0
𝐿𝐸𝑆𝑆
𝑖,𝑡,𝑠,𝑠𝑒 = 𝑡ℎ 𝑡ℎ 𝑡ℎ
𝑑𝑐ℎ
−𝛼𝑛4 (𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝑐ℎ
) − 𝛽𝑛4 𝑄𝑖,𝑡,𝑠,𝑠𝑒 − 𝛾 4 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 ≤ 0
4 𝑡ℎ𝑑𝑐ℎ 𝑐ℎ 4 𝑡ℎ
𝐸𝑆𝑆 4 𝑡ℎ
𝐸𝑆𝑆,𝑚𝑎𝑥
{ 𝛼𝑛 (𝑃𝑖,𝑡,𝑠,𝑠𝑒 − 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ) − 𝛽𝑛 𝑄𝑖,𝑡,𝑠,𝑠𝑒 − 𝛾 𝑆𝑖 ≤0
,𝑛 = 1,2, … , 𝑛𝑝𝑤 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.86)
97
𝐸𝑆𝑆 𝑆𝑂𝐶𝑖0 𝐸𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 + (𝜂𝑖𝐸𝑆𝑆 𝑃𝑖,𝑡,𝑠,𝑠𝑒
𝑐ℎ 𝑑𝑐ℎ
− 𝑃𝑖,𝑡,𝑠,𝑠𝑒 )∆𝑡, 𝑡 = 1
𝐸𝑖,𝑡,𝑠,𝑠𝑒 ={ 𝐸𝑆𝑆 𝐸𝑆𝑆 𝑐ℎ 𝑑𝑐ℎ
𝐸𝑖,𝑡−1,𝑠,𝑠𝑒 + (𝜂𝑖 𝑃𝑖,𝑡,𝑠,𝑠𝑒 − 𝑃𝑖,𝑡,𝑠,𝑠𝑒 )∆𝑡, 𝑡 ≥ 2
,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.87)
𝐷𝑜𝐷𝑖𝑚𝑎𝑥 𝐸𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 ≤ 𝐸𝑖,𝑡,𝑠,𝑠𝑒

𝐸𝑆𝑆
≤ 𝐸𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒
,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.88)
𝐸𝑆𝑆
𝐸𝑖,𝑡=𝑁𝑇 ,𝑠,𝑠𝑒
≥ 𝑆𝑂𝐶𝑖0 𝐸𝑖𝐸𝑆𝑆,𝑠𝑖𝑧𝑒 ,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.89)
−𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 (1 − 𝑏𝑡,𝑠,𝑠𝑒
𝑚 𝑃𝐶𝐶
) ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 (1 − 𝑏𝑡,𝑠,𝑠𝑒
𝑚
)
,∀𝑖 ∈ Ω𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.90)
𝜒𝑖 𝑇𝑅,𝑚𝑖𝑛 𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 (1 − 𝑏𝑡,𝑠,𝑠𝑒

𝑚 𝑃𝐶𝐶
) ≤ 𝑄𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝜒𝑖 𝑇𝑅,𝑚𝑎𝑥 𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 (1 − 𝑏𝑡,𝑠,𝑠𝑒
𝑚
)
,∀𝑖 ∈ Ω𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.91)
𝑃𝐶𝐶
𝛼𝑛1 𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝑃𝐶𝐶
+ 𝛽𝑛1 𝑄𝑖,𝑡,𝑠,𝑠𝑒 − 𝛾 1 𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 ≤ 0
𝑃𝐶𝐶
−𝛼𝑛1 𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝑃𝐶𝐶
+ 𝛽𝑛1 𝑄𝑖,𝑡,𝑠,𝑠𝑒 − 𝛾1 𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 ≤ 0
𝐿𝑃𝐶𝐶
𝑖,𝑡,𝑠,𝑠𝑒 = 𝑡ℎ 𝑡ℎ 𝑡ℎ
𝑃𝐶𝐶
−𝛼𝑛4 𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝑃𝐶𝐶
− 𝛽𝑛4 𝑄𝑖,𝑡,𝑠,𝑠𝑒 − 𝛾 4 𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 ≤ 0
4 𝑡ℎ
𝑃𝐶𝐶 4 𝑡ℎ
𝑃𝐶𝐶 4 𝑡ℎ
𝑇𝑅,𝑚𝑎𝑥
{ 𝛼𝑛 𝑃𝑖,𝑡,𝑠,𝑠𝑒 − 𝛽𝑛 𝑄𝑖,𝑡,𝑠,𝑠𝑒 − 𝛾 𝑆𝑖 ≤0
,𝑛 = 1,2, … , 𝑛𝑝𝑤 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑖 ∈ Ω𝑃𝐶𝐶 , , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.92)
𝑃𝐶𝐶
−𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝑡𝑔(cos−1 𝑃𝐹𝑖𝑇𝑅,4𝑡ℎ ) ≤ 𝑄𝑖,𝑡,𝑠,𝑠𝑒
𝑃𝐶𝐶 𝑃𝐶𝐶
≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝑡𝑔(cos −1 𝑃𝐹𝑖𝑇𝑅,1𝑠𝑡 )
,∀𝑖 ∈ Ω𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.93)
𝑚 𝑚
𝑉𝑖,𝑡,𝑠,𝑠𝑒 (1 − 𝑏𝑡,𝑠,𝑠𝑒 ) = 𝑉 𝐺𝐶,𝑠𝑝𝑒𝑐 (1 − 𝑏𝑡,𝑠,𝑠𝑒 )
,∀𝑖 ∈ Ω𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.94)
𝑚
𝛿𝑖,𝑡,𝑠,𝑠𝑒 (1 − 𝑏𝑡,𝑠,𝑠𝑒 )=0 ,∀𝑖 ∈ Ω𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.95)
The inequality (5.63) represents the general constrain that guarantees the voltage magnitude for
difference individual buses within magnitude limits. Eq. (5.64) guarantees load curtailment during IS
mode. The microgrid model is obtained using the power flow model given by (5.65)-(5.69).
Constraints (5.65) and (5.66) are the linear version for the node-based power flow equation that refers
to the power injected to the 𝑖 𝑡ℎ bus of the microgrid [3]. It is worth noting here that, (5.65) and (5.66)
have been generally formulated for all buses. Thus, the variables representing the power exchange
and generation (i.e., by DDG, WT, PV, ESS) that are not relevant to the power flow equations written
98
for bus 𝑖 can be disregarded by setting the upper bound for these variables to zeros. Next, eqs. (5.67)
and (5.68) are the linear version defining the active and reactive power flowing along a branch 𝑘 in
the microgrid. Where, 𝑖(𝑘) and 𝑗(𝑘) denote buses i and j located at the upstream and downstream of
branch 𝑘, respectively. Finally, (5.69) is the piecewise linearization of the quadratic constraint
limiting the line ampacity.
The DDG units are modeled using the steady state model (5.70)-(5.79). Figure 5. 6 shows the steady
state model employed for representing operation constraints for the DDG units with droop/2nd .
Details of the derivation of the DDG unit operation constraints can be found in [3]. Constraints (5.70)
and (5.71) define the active and reactive power generated by the DDG units during IS or GC mode.
𝑚
These constraints are composed of two terms associated with the binary input 𝑏𝑡,𝑠,𝑠𝑒 . The first term
of these constraints refers to the operation of the DDG unit under droop/2nd control during IS mode.
The second terms represent the operation of the DDG unit in PQ mode during grid connected mode.
As one may observe, the activation of the terms in (5.70) and (5.71) are conditioned to the binary
𝑚
input 𝑏𝑡,𝑠,𝑠𝑒 status. Eq. (5.72) guarantees the voltage of the DDG bus restoration to the nominal
secondary setting voltage in IS mode. Eq. (5.73) defines the apparent power equal to the DDG
allocated unit rated size. In this work, loads are considering to be shared proportionally to the capacity
of the DDG units [26],[54]. Thus, the active and reactive droop gain can be calculated via the
capacity-based model as expressed in (5.74) and (5.75). Eqs. (5.76) and (5.77) establish minimum
and maximum boundary limits to the active and reactive power.
MICROGRID
fbus
PLi , QLi M1
kW
kVAR
BUSj
(PCCj) DDG
upstream downstream
Hierachical control Feasible operation control

PD DG j QDDG
QD DG j M1 SDDGmax
SRF-PLL
PDDG
|Vj| δMG
Droop
control action
PCCi
Secondary DDG
control action
Fiseable operation
GND Droop ch s region
FIGURE 5. 6: Steady state model for DDG unit operate with 𝐝𝐫𝐨𝐨𝐩/𝟐𝐧𝐝 in IS mode and their
feasible operation region for both GC and IS mode
99
Constraint (5.78) guarantees that the relationship between the active and reactive power of a DDG
unit is greater than the minimum set power factor. Finally (5.79) represents the set of piecewise
linearization that ensures the operation of the DDG unit respecting its rated capacity 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 .
Constraint (5.80) and (5.81) represent the power outputs of the WTs and PVs considering their actual
𝑊𝑇 𝑃𝑉
MPPT power 𝑃𝑖,𝑡,𝑠,𝑠𝑒 and 𝑃𝑖,𝑡,𝑠,𝑠𝑒 . ESS constraints are depicted in (5.82)-(5.89). For this, in this work,
ESS unit are operated in PQ mode via its power electronic inverter interfaced (5.82). Nevertheless,
with the power electronic interface the ESS can be operated with droop/2nd control similar to the
work in [3]. Constraint (5.83)-(5.84) guarantee that the ESS units operation are restricted within the
feasible operation region. In these constraints, the binary variable 𝑏𝐸𝑆𝑆
𝑖,𝑡,𝑠,𝑠𝑒 is introduced in (5.83) and
(5.84) to avoid simultaneous charging and discharging states. Eq. (5.85) ensures that the reactive
power delivered by the ESS units respect their maximum and minimum reactive power allowable
setting. As the ESS unit is connected through an inverter, (5.86) defines the piecewise linearization
representing the quadratic constrain for limiting the output to the maximum rated power denote by
𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 . In (5.87) update SOC for the ESS are obtained for each time segment based on the
charging/discharging energy flows, as well as considering the round-trip efficiency 𝜂𝑖𝐸𝑆𝑆 [85], [84].
Constraint (5.88) guarantees that the energy stored in the ESS is restricted by its rated capacity and
minimum energy storage setting. Finally (5.89) ensures that the energy stored in the ESS is available
at the end of the planning horizons and ready for the next days.
Constraints for the power exchange between the microgrid and the main grid at the substation are
given by (5.32)-(5.37). The constraints (5.90)-(5.93) guarantee the power exchange is restricted by
transformers’ feasible operation. First, (5.90) and (5.91) limit the active and reactive power exchange
𝑚
respecting to the transformer rated power. As well, the introduction of 𝑏𝑡,𝑠,𝑠𝑒 in (5.90) and (5.91)
allows for deactivating these constraints during IS mode. Then, (5.92) is the piecewise linearization
representing the quadratic constrain for limiting the active and reactive power exchange under the
transformer rated capacity. The operation for transformer limits above the minimum power factor
setting is given by (5.93). Finally (5.94) and (5.95) refer to the grid connected power flow conditions
for voltage magnitude and phase angle at the PCC MG imposed by the main grid during grid connected
mode.
The piecewise constraint coefficients for (5.69), (5.79), (5.86) and (5.92) can be calculated using
(5.96)-(5.99). Details on the derivation of piecewise constrains coefficients for the quadratic
constraints can be found in [3].
100
𝛼𝑛1 = cos (𝜑𝑛1 + (𝑛1 1 ,4
− 1) 𝜑𝑠𝑒𝑙 ⁄𝑛1𝑝𝑤,4 ) ,𝑛 = 1,2, … , 𝑛𝑝𝑤 (5.96)

𝛽𝑛1 = sin (𝜑𝑛1 + (𝑛1 1 ,4
− 1) 𝜑𝑠𝑒𝑙 ⁄𝑛1𝑝𝑤,4 ) ,𝑛 = 1,2, … , 𝑛𝑝𝑤 (5.97)
𝑠𝑡 ,4𝑡ℎ 1 ,4 𝑠𝑡 𝑡ℎ 𝑠𝑡 𝑡ℎ
𝛾1 = cos (𝜑𝑠𝑒𝑙 ⁄2𝑛1𝑝𝑤,4 ) ,𝑛 = 1,2, … , 𝑛𝑝𝑤 (5.98)
𝑠𝑡 ,4𝑡ℎ 𝑠𝑡 𝑡ℎ 𝑠𝑡 𝑡ℎ
𝜑01 1 ,4
= 𝜑𝑠𝑒𝑙 ⁄2𝑛1𝑝𝑤,4 ,𝑛 = 1,2, … , 𝑛𝑝𝑤 (5.99)
5.4.2.2 Decision variables for Inner Level
The decision variables for the MILP model seeking to optimize microgrids’ operation are summarized
in Table 5. 2.
TABLE 5. 2: Decision variables for MILP model for microgrid optimal operation
Variables Set
𝑴𝑰𝑳𝑷 Type ∀𝑡 ∈ , ∀𝑠 ∈ ∀𝑠𝑒 ∈ ∀𝑖, 𝑘 ∈
𝑿
′ ′′
𝑉𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω𝑠 ∪ Ω𝑠 Ω𝑠𝑒 Ω𝐵𝑈𝑆
𝛿𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω′𝑠 ∪ Ω′′
𝑠 Ω𝑠𝑒 Ω𝐵𝑈𝑆
𝑐𝑢𝑟𝑡 ′ ′′
𝑏𝑖,𝑡,𝑠,𝑠𝑒 ℤ 𝛺𝑇 Ω𝑠 ∪ Ω𝑠 Ω𝑠𝑒 Ω𝐵𝑈𝑆
𝐷𝐷𝐺 ′ ′′
𝑃𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω𝑠 ∪ Ω𝑠 Ω𝑠𝑒 Ω𝐷𝐷𝐺
𝐷𝐷𝐺 ′ ′′
𝑄𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω𝑠 ∪ Ω𝑠 Ω𝑠𝑒 Ω𝐷𝐷𝐺
𝑣 ′ ′′
𝑢𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω𝑠 ∪ Ω𝑠 Ω𝑠𝑒 Ω𝐷𝐷𝐺
𝑊𝑇,𝑢𝑠𝑒 ′ ′′
𝑃𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω𝑠 ∪ Ω𝑠 Ω𝑠𝑒 Ω𝑊𝑇
𝑃𝑉,𝑢𝑠𝑒 ′ ′′
𝑃𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω𝑠 ∪ Ω𝑠 Ω𝑠𝑒 Ω𝑃𝑉
𝑑𝑐ℎ
𝑃𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω′𝑠 ∪ Ω′′
𝑠 Ω𝑠𝑒 Ω𝐸𝑆𝑆
𝑐ℎ
𝑃𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω𝑠 ∪ Ω′′
′
𝐸𝑆𝑆
𝑄𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω′𝑠 ∪ Ω′′
𝐸𝑆𝑆
𝐸𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω′𝑠 ∪ Ω′′
𝑏𝐸𝑆𝑆
𝑖,𝑡,𝑠,𝑠𝑒 ℤ 𝛺𝑇 Ω′𝑠 ∪ Ω′′
𝑃𝐶𝐶
𝑃𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω′𝑠 ∪ Ω′′
𝑠 Ω𝑠𝑒 Ω𝑃𝐶𝐶
𝑃𝐶𝐶
𝑄𝑖,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω𝑠 ∪ Ω′′
′
𝑠 Ω𝑠𝑒 Ω𝑃𝐶𝐶
𝑏𝑟
𝑃𝑘,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω𝑠 ∪ Ω′′
′
𝑠 Ω𝑠𝑒 Ω𝐿𝐼𝑁
𝑏𝑟
𝑄𝑘,𝑡,𝑠,𝑠𝑒 ℝ 𝛺𝑇 Ω𝑠 ∪ Ω′′
′
𝑠 Ω𝑠𝑒 Ω𝐿𝐼𝑁
5.5 Stage III: Optimal Operation based on Deep Neural Network (DNN)
learning
The detailed modeling of the multiple components involved in microgrids for ensuring adequate
operating conditions during grid-connected and isolated modes, make it unfeasible to solve this
101
problem using a MILP formulation due to the stochastic nature of the problem, and consequently, the
implied high computational burden. In this sense, seeking to overcome these limitations while not
reducing the modeling complexity in order to ensure feasible resilient solutions, a novel methodology
using DNN technique is proposed. For this, effective training samples for the DNN models are
necessary. These samples are composed of 1) the set of input samples for combinations of switch
allocation and resources; and 2) the set of targets for the LOPF dependent indices used for determining
the multi-objective functions of the outer level. To obtain this data, the LOPF model depicted in the
Section 5.3 is used considering microgrid formation indicators-based N-K contingencies level. Here,
given that the LOPF is simulated for set of sample inputs belonging to Ω𝑠 , Ω′′
𝑠 and Ω𝑠𝑒 , which are
independent to each other, parallel computing can be employed as shown in Figure 5. 7. For this, any
third-party solution based on well-knows multi-CPU cores or multi-GPU cores can be applied.
A general outline of this process based on the stochastic DL-LOPFAC approach considering
microgrids operation and control is presented in Figure 5. 7. Details for obtaining the set of input
samples, as well as the targets provided by the LOPF model are described in the following section.
5.5.1 Set of Sample Inputs and Targets
The set of input samples is a combination of possible allocations for the DDGs, WTs, PVs, ESSs and
ISWs, where 𝐈𝐍 𝐝𝐧𝐧 denotes the matrix including a set of sample 𝑠𝑝 that can be expressed by (5.100)-
(5.101).
𝐈𝐍 𝐝𝐧𝐧 = [𝑰𝑵1𝒅𝒏𝒏 ; 𝑰𝑵𝒅𝒏𝒏 𝒅𝒏𝒏

2 ; … ; 𝑰𝑵𝑠𝑝 ] ,𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.100)
𝑟𝑜𝑢𝑛𝑑(𝒓𝑑𝑛𝑛
𝑎,𝑠𝑝 )
𝑃𝐷𝐷𝐺,𝑚𝑎𝑥 𝑠𝑝 ∈ Ω𝑆𝐴𝑀
𝑟𝑜𝑢𝑛𝑑 (𝒓𝑑𝑛𝑛
𝑏,𝑠𝑝 .
𝐷𝐷𝐺
𝐷𝐷𝐺 ) 𝑃𝑠𝑡𝑒𝑝
𝑃𝑠𝑡𝑒𝑝
𝑎 ∈ Ω𝑐𝑎𝑛
𝐼𝑆𝑊
𝑃𝑊𝑇,𝑚𝑎𝑥 𝑏 ∈ Ω𝑐𝑎𝑛
𝑰𝑵𝒅𝒏𝒏 = 𝑟𝑜𝑢𝑛𝑑 (𝒓𝑑𝑛𝑛
𝑐,𝑠𝑝 . 𝑃𝑠𝑡𝑒𝑝
𝑊𝑇
𝑊𝑇 ) 𝑃𝑠𝑡𝑒𝑝 , 𝐷𝐷𝐺
(5.101)
𝑠𝑝
𝑐 ∈ Ω𝑐𝑎𝑛
𝑊𝑇
𝑃𝑃𝑉,𝑚𝑎𝑥
𝑑,𝑠𝑝 . 𝑃𝑉
𝑃𝑉
) 𝑃𝑠𝑡𝑒𝑝 𝑑 ∈ Ω𝑐𝑎𝑛
𝑃𝑉
𝑃𝑠𝑡𝑒𝑝
𝑒 ∈ Ω𝑐𝑎𝑛
𝐸𝑆𝑆
𝑃𝐸𝑆𝑆,𝑚𝑎𝑥 𝐸𝑆𝑆
𝑒,𝑠𝑝 . 𝐸𝑆 ) 𝑃𝑠𝑡𝑒𝑝
[ 𝑃𝑠𝑡𝑒𝑝 ]
The vector 𝑰𝑵𝒅𝒏𝒏

𝑠𝑝 is dependent on the candidate location of the ISW and resources, where 𝑛𝑓𝑒𝑎𝑡
denotes the number of features for a single DNN training sample input, in this case 𝑛𝑓𝑒𝑎𝑡 = 𝑛𝑑𝑑𝑔 +
𝑛𝑤𝑡 + 𝑛𝑝𝑣 + 𝑛𝑒𝑠𝑠 + 𝑛𝑖𝑠𝑤 , 𝒓𝑑𝑛𝑛
𝑠𝑝 is a uniform random number between [0,1]. In this work, the uniform
random is generated one time as a matrix with size 𝑛𝑓𝑒𝑎𝑡 × 𝑛𝑠𝑎𝑚 representing a whole Ω𝑆𝐴𝑀 training
102
sample inputs before being classified into subset according to the candidate location of switch and
resources as seen in (5.101). In this way, it guarantees that the generated random numbers cover all
possible configuration in the space 𝑛𝑓𝑒𝑎𝑡 × 𝑛𝑠𝑎𝑚 .
From
From stage 1
stage 2a
1) candidate location and size for resources C
2) candidate location for switches
Generate samples inputs Identify outer

for using uniform level LOPF
distribution dependent indices
Set of
From From
Sample stage 1 stage 1
Inputs A B
( SAM) START
Read data
AVnew
forming microgrid
formation indicator
for scenario {s,se}
LOPFsev{1,1} LOPFnor{1,1}
...
...
LOPFsev{Ns ,1 LOPFnor{Ns ,1
...
...
LOPFsev{Ns ,Nse} LOPFnor{Ns ,Nse}

LOPF for severe LOPF for severe
situation scenarios situation scenarios
Store spth LOPF solution

i.e., matrix of variables
Parallel sp>nsam no
computing LOPF
Set of LOPF solutions ( SAM)
Determine LOPF
dependent indices
Set of Samples target ( SAM)
input NN output
DNN trainning
Adjust
hyper-parameters
Criteria
satisfied?
yes
Trainned DNN models

net1 net2 ... netn
END
FIGURE 5. 7: Proposed optimal operation for hierarchical microgrids based on deep-learning

technique
103
The set of targets are identified by outer level’s indices dependent on the decision variables of the
LOPF problem. In this case, from (5.21)-(5.46) the identified indices are 𝑅𝐸𝑆𝐼 𝑂𝐿𝐸𝑉 , 𝐴𝑂𝑀𝐶 𝐷𝐷𝐺 ,
𝐴𝐸𝑀𝐼𝑆 𝐷𝐷𝐺 and 𝐴𝐸𝑀𝐼𝑆 𝐺𝑅𝐼𝐷 , where 𝑇𝐺𝑠𝑝,𝑛 denotes the DNN targets 𝑛 corresponding to LOPF
solution sample 𝑠𝑝, the identification targets for training the DNN model can be given by (5.102)-
(5.104).
𝑇𝐺𝑠𝑝,1 = 𝑅𝐸𝑆𝐼 𝑂𝐿𝐸𝑉 ,∀𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.102)
𝐴𝑂𝑀𝐶 𝐷𝐷𝐺
365 𝜎𝑓𝑢𝑒𝑙 𝐷𝐷𝐺 )
𝑆 ( +𝜎𝑂&𝑀
4 𝑏𝑎𝑠𝑒 𝜂𝐷𝐷𝐺
𝑤𝑖𝑠𝑜 ∑ 𝐷𝐷𝐺
𝑇𝐺𝑠𝑝,2 = 𝑜𝑟 = ∑∀𝑠𝑒∈Ω𝑠𝑒 ∑∀𝑠∈Ω𝑠(𝜋𝑠,𝑠𝑒 ∀𝑖∈Ω𝐷𝐷𝐺 ∑∀𝑡 ∈ Ω𝑇 𝑃𝑖,𝑡,𝑠,𝑠𝑒 )
𝐴𝐸𝑀𝐼𝑆 𝐷𝐷𝐺
365
{ 𝑆 𝐾𝐷𝐷𝐺
4 𝑏𝑎𝑠𝑒 𝐶𝑂2
,∀𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.103)
𝐴𝐸𝑀𝐼𝑆 𝐺𝑅𝐼𝐷 𝑤𝑖𝑠𝑜 ∑ 𝐺𝑅𝐼𝐷+

𝑇𝐺𝑠𝑝,3 = 365 𝐺𝑅𝐼𝐷
= ∑∀𝑠𝑒∈Ω𝑠𝑒 ∑∀𝑠∈Ω𝑠(𝜋𝑠,𝑠𝑒 ∀𝑖∈Ω𝐷𝐷𝐺 ∑∀𝑡 ∈ Ω𝑇 𝑃𝑡,𝑠,𝑠𝑒 )
𝑆𝑏𝑎𝑠𝑒 𝐾𝐶𝑂2
4
,∀𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.104)
The first target (5.102), 𝑅𝐸𝑆𝐼 𝑂𝐿𝐸𝑉 , is dependent to LOPF’s decision variable 𝑏𝑐𝑢𝑟𝑡
𝑖,𝑡,𝑠,𝑠𝑒 . It is responsible
for the training of DDN model for accounting 𝑏𝑐𝑢𝑟𝑡

𝑖,𝑡,𝑠,𝑠𝑒 of the inner level. This allow the trained DNN
model enable to predict the resilience index value which is corresponding to the first objective
function as discussed in section 5.4. The second target (5.103) is related to 𝐴𝑂𝑀𝐶 𝐷𝐷𝐺 and 𝐴𝐸𝑀𝐼𝑆 𝐷𝐷𝐺
𝐷𝐷𝐺
indices dependent to the LOPF’s decision variables 𝑃𝑖,𝑡,𝑠,𝑠𝑒 . It characterizes both indices shared by
𝐷𝐷𝐺
the same variable i.e., 𝑃𝑖,𝑡,𝑠,𝑠𝑒 and leading to the possibility to predict only a single value for
determining these two indices. As can see in (5.103), 𝑇𝐺𝑠𝑝,2 can be calculated via 𝐴𝑂𝑀𝐶 𝐷𝐷𝐺 , or
𝐷𝐷𝐺
𝐴𝐸𝑀𝐼𝑆 𝐷𝐷𝐺 , as those indices are dependent to the same 𝑃𝑖,𝑡,𝑠,𝑠𝑒 . In this case, the prediction results
according to 𝑇𝐺𝑠𝑝,2 can be used for re-calculating 𝐴𝑂𝑀𝐶 𝐷𝐷𝐺 and 𝐴𝐸𝑀𝐼𝑆 𝐷𝐷𝐺 respecting their
corresponding multiplication constant requirement. These re-calculated provide the predicted results
for 𝐴𝑂𝑀𝐶 𝐷𝐷𝐺 and 𝐴𝐸𝑀𝐼𝑆 𝐷𝐷𝐺 which is necessary for determine the second/third objective function
for the outer level (i.e., 𝑇𝐶 𝑂𝐿𝐸𝑉 and 𝐴𝐸𝑀𝐼𝑆 𝑂𝐿𝐸𝑉 ) accordingly. Finally (5.104) corresponding to
𝐺𝑅𝐼𝐷+
𝐴𝐸𝑀𝐼𝑆 𝐺𝑅𝐼𝐷 which it dependent to 𝑃𝑖,𝑡,𝑠,𝑠𝑒 , leading to the representation of the predicted value of the
trained DNN model for the annualized CO2 emission by the power imported from the main grid during
GC mode, similar to (5.45) in section 5.4.
104
5.5.2 Microgrid Formation
The microgrid formation is the intermediate step within the stochastic optimal operation level. This
step delivers the necessary indicators for optimal operation level representing the microgrids
formation due to N-K contingencies level. These indicators are the matrices and vectors that represent
the time-coupling topologies of test system during severe scenarios Ω′′
𝑠 . The methodology for
delivering the microgrid formation indicators respecting to the LOPF problem is detailed illustrated
in Figure 5. 8.
Set of input START

samples Import line data R, X, fb, tb
fb, tb, brnum
ISW site(sp),
DDG site(sp), [LFa] and [BFa] [LFa]
DDG size(sp) floct ISW site(sp)
Path finder algorithm Determine MGtset Generate Ytbus

Generate Graph and
Adjacency matrix for yes t>NT t=t+1 t>NT t=t+1
netwok tropology
yes
Generate vector node TGnode Determine btm
Determine t>NT t=t+1

PATHbus and PATHbr
yes
bus br Determine
Forming FA and FA DDG site(sp)
ftbus & DDG tdown
no brifail>nbr t>NT t=t+1

yes
DDG site(sp)
Determine bustne
Forming BFat and LFat DDG size(sp)
no s>NT t>NT t=t+1

yes
floct busne Ybus
WTdown bm
PVdown fbus
Generate ESS down DDG down
floct
indicators set for
AVI-IIs,se {s,se}
A
From END
stage 1
LOPFsev{s,se}
FIGURE 5. 8: Flowchart for generating the microgrid indicator matrices and vectors for time
𝒕 ∈ 𝛀𝑻 for scenarios s of season se
105
5.5.2.1 Fault Influence Path
First, the affected branch and bus path finder is proposed for searching the path(s) from the original
fault location until the end branch (terminal edge) and end node (leaves node) respecting the graph
representation of the network. This search algorithm is deployed in order to automatically search the
path(s) whenever a single fault is located in the network. The respective step-by-step description of
the algorithm solution is detailed in the following sections. On the other hand, more detail exemplifies
the determination of the fault influence path can be found in Appendix C.
a) Step 1: Representation of the System Topology
Let the reference system topology be represented by the undirected graph 𝐆𝐛𝐮𝐬 = (𝑵𝑫𝒃𝒖𝒔 , 𝑬𝑫𝒃𝒓 ),
where 𝑵𝑫𝒃𝒖𝒔 are the set containing the vertices or nodes representing the buses in the reference
system, while 𝑬𝑫𝒃𝒓 are the edge set representing the branch connected between two adjacent nodes
in the test system. In this case, the adjacency matrix of 𝐆𝐛𝐮𝐬 is the 𝑛𝑏𝑢𝑠 × 𝑛𝑏𝑢𝑠 matrix denote by
𝑎𝑑𝑗 𝑎𝑑𝑗
𝐌 𝐚𝐝𝐣 = [𝑀𝑖𝑗 ] with 𝑀𝑖𝑗 defined by (5.105) [105].
𝑎𝑑𝑗 1, if 𝑁𝐷𝑖𝑏𝑢𝑠 𝑁𝐷𝑗𝑏𝑢𝑠 ∈ 𝑬𝑫𝒃𝒓

𝑀𝑖𝑗 ={ (5.105)
0, otherwise
𝑎𝑑𝑗
As 𝐆𝐛𝐮𝐬 is an undirected graph, the adjacency matrix is the symmetrical square matrix with 𝑀𝑖𝑗 =
𝑎𝑑𝑗
𝑀𝑗𝑖 . Here it is important to note that in case there is a reconfiguration in the reference system, 𝐆𝐛𝐮𝐬
and 𝐌 𝐚𝐝𝐣 should be updated accordingly.
b) Step 2: Determination of the Target Nodes 𝑻𝑮𝒏𝒐𝒅𝒆
Based on the graph 𝐆𝐛𝐮𝐬 , the shortest path by Dijkstra’s algorithm is adopted for determining the
path(s), or road map(s), between the branch fail and the target node(s), where 𝑓𝑏(𝑏𝑟 𝑓𝑎𝑖𝑙 ) denotes the
start node for the shortest path, i.e. fault location, while the end of the shortest path are the target
nodes including root node and leave nodes. More detail about the Dijkstra’s algorithm can be found
in [105],[106].
The target nodes can be identified using the adjacency matrix 𝐌 𝐚𝐝𝐣 , determining the row of 𝐌 𝐚𝐝𝐣 with
𝑛𝑏𝑢𝑠 𝑎𝑑𝑗
summation equal to one. In this sense, let 𝑖 and 𝑗 denote the row and column of 𝐌 𝐚𝐝𝐣 , ∑𝑗=1 𝑀𝑖,𝑗 =
1 indicates that node 𝑖 is an end node with no connection elsewhere.
106
c) Step 3: Determination of the path(s)
Once the target nodes are identified, Dijkstra’s algorithm is employed for determining the shortest
path between the start node 𝑓𝑏(𝑏𝑟 𝑓𝑎𝑖𝑙 ) and the target nodes 𝑇𝐺𝑖𝑛𝑜𝑑𝑒 . For this, considering the system
topology information available in the 𝐆𝐛𝐮𝐬 matrix, where the weight of all branches are set to one,
given that any connection between two nodes is established by a unique path for a radial distribution
system.
The Dijkstra’s algorithm solution (𝐷𝐼𝐽𝐾) identifying the shortest path between two nodes,
respectively 𝑓𝑏(𝑏𝑟 𝑓𝑎𝑖𝑙 ) and 𝑇𝐺𝑖𝑏𝑢𝑠 , can be expressed by eq. (5.106).
𝒑𝒂𝒕𝒉𝒃𝒖𝒔
𝑖 (𝑓𝑏(𝑏𝑟
𝑓𝑎𝑖𝑙
), 𝑇𝐺𝑖𝑏𝑢𝑠 ) = 𝐷𝐼𝐽𝐾(𝑓𝑏(𝑏𝑟 𝑓𝑎𝑖𝑙 ), 𝑇𝐺𝑖𝑏𝑢𝑠 ) ,∀𝑖 ∈ Ω𝑛𝑜𝑑𝑒
𝑇𝐺 (5.106)
Next, the branch path denoted by 𝑝𝑎𝑡ℎ𝑖𝑏𝑟 vector can be obtained identifying the edges located within
the solution path 𝑓𝑏(𝑏𝑟 𝑓𝑎𝑖𝑙 ) to 𝑇𝐺𝑖𝑏𝑢𝑠 . Following, vectors 𝑝𝑎𝑡ℎ𝑖𝑏𝑢𝑠 and 𝑝𝑎𝑡ℎ𝑖𝑏𝑟 belonging to Ω𝑛𝑜𝑑𝑒
𝑇𝐺
are stored in matrices 𝐏𝐀𝐓𝐇 𝐛𝐮𝐬 and 𝐏𝐀𝐓𝐇 𝐛𝐫, respectively. These matrices are responsible for
determining the fault influenced branch and bus matrices, being detailed elaborated in the following
section.
5.5.2.2 Fault Influenced Branch and Bus Matrices
Following, once affected paths are determined, clearing procedures must be established to
successfully isolate the fault and ensure the system resilience. In this sense, the representation of
affected bus(es) and branch(es) by the respective fault(s) must be performed to enable planning
algorithms solution. These representations are respectively held by matrices 𝐋𝐅𝐚 and 𝐁𝐅𝐚, describing
affected bus(es) and branch(es). In addition, the establishment of rules to ensure the adequate
allocation of candidate isolated switches able to isolate the upstream and downstream of a branch
fault is fundamental [107][108]. For this, the following rules are established:
• If the fault is found on the branch equipped with ISW, the fault stays in the middle of the
branch and have been cleared by this ISW.
• If the fault is located on the branch without ISW, the fault in this case is cleared by the nearby
ISWs located along the searching paths.
A detailed description of bus and branch affected matrices is presented in the following sections.
107
a) Branch Affected Matrix (𝑳𝑭𝒂)
The branch, or line, affected matrix 𝐋𝐅𝐚 is the binary matrix assembling all binary vectors 𝑳𝑭𝒂𝑡 with
𝑡 belong to Ω𝑇 , where an online branch is represented by “1”, i.e., non-affected branch, and affected
branches by the fault(s) are represented by “0”. For determining this matrix, first, the binary matrix
denoted by 𝐅𝐀𝐛𝐫 is required. 𝐅𝐀𝐛𝐫 is the binary square matrix 𝑛𝑏𝑟 × 𝑛𝑏𝑟 , in which its 𝑖 𝑡ℎ single row
𝑓𝑎𝑖𝑙
describes the set of affected branches corresponding to the location of a single fault 𝑏𝑟𝑖 where 𝑖 ∈
Ω𝐿𝐼𝑁 . Once matrix 𝐅𝐀𝐛𝐫 is obtained, a generic 𝑖 𝑡ℎ fault located at a branch “a”, i.e., 𝑓𝑙𝑜𝑐𝑖,𝑡 = 𝑎, due
to the severe event at time “𝑡”, allows for the determination of 𝑳𝑭𝒂𝑡 by the row “𝑎” of 𝐅𝐀𝐛𝐫 .
Moreover, 𝐅𝐀𝐛𝐫 is useful for manipulating the elements of 𝑳𝑭𝒂𝒕 in case that multiple faults are
featured in the test system (i.e., 𝑓𝑙𝑜𝑐𝑖,𝑡 ∈ Ω𝑓𝑙𝑜𝑐 ). To this end. the location of the fault(s) can be
𝐈−𝐈𝐈
identified by looking up on time segment 𝑡 of 𝐀𝐕𝑠,𝑠𝑒 .
If there is more than a single fault located in the system. Let, 𝒇𝒍𝒐𝒄𝑡 be a vector containing the faults
I−II
locations, where its elements can be identified using 𝐀𝐕𝑠,𝑠𝑒 . 𝑳𝑭𝒂𝒕 at time 𝑡 can be determined by
looking up the multiple rows of 𝐅𝐀𝐛𝐫 considering the correspondent element number of vectors 𝒇𝒍𝒐𝒄𝑡
as described in (5.107), where ∏ denotes the Hadamard multiplication.
𝑳𝑭𝒂𝑡 = ∏𝑖∈Ω𝑓𝑙𝑜𝑐[∘ 𝐅𝐀𝐛𝐫 (𝑓𝑙𝑜𝑐𝑖 , : )] (5.107)
Finally, 𝐋𝐅𝐚 is obtained by repeating the same methodology for 𝑳𝑭𝒂𝑡 for 𝑡 ∈ Ω𝑇 . Here one must note
that, if no fault(s) occur at time 𝑡, especially in the normal scenarios, i.e., Ω𝑠 , the elements of 𝑳𝑭𝒂𝑡
can be simply set by the vector constrained element of ‘1’. More detail elaborates the determination
of 𝑳𝑭𝒂𝑡 until 𝐋𝐅𝐚 can be found in Appendix C.
b) Bus Affected Matrix (𝑩𝑭𝒂)
Similar to the branches analysis, buses located within the fault zone bounded by the ISWs must be
deactivated. In this sense, to represent this information, the binary matrix denote by 𝐁𝐅𝐚 containing
the bus(es) affected is required, where the 𝐁𝐅𝐚 is an assembly of the binary vectors 𝑩𝑭𝒂𝑡 where 𝑡 ∈
Ω𝑇 , with term(s) equal to “1”, corresponding to an online bus, and “0” for those affected by the fault(s)
and deactivated.
Following the binary matrix denoted by 𝐅𝐀𝐛𝐮𝐬 is required. 𝐅𝐀𝐛𝐮𝐬 is a non-square matrix due to 𝑛𝑏𝑢𝑠 =
𝑛𝑏𝑟 + 1, where the 𝑖 𝑡ℎ single row of 𝐅𝐀𝐛𝐮𝐬 describes the set of affected buses corresponding to the
𝑓𝑎𝑖𝑙
location of a single fault location 𝑏𝑟𝑖 given 𝑖 ∈ Ω𝐿𝐼𝑁 . Once this matrix 𝐅𝐀𝐛𝐮𝐬 obtained, the 𝑩𝑭𝒂𝑡
108
can be determined by looking up to the row “𝑎” of 𝐅𝐀𝐛𝐮𝐬. Similar to 𝐅𝐀𝐛𝐫 , 𝐅𝐀𝐛𝐮𝐬 can be used for
determining 𝑩𝑭𝒂𝑡 in case the case there is a multi-fault in the network i.e., 𝑓𝑙𝑜𝑐𝑖,𝑡 ∈ Ω𝑓𝑙𝑜𝑐 .
𝑓𝑎𝑖𝑙
For determining 𝐅𝐀𝐛𝐮𝐬, first 𝐏𝐀𝐓𝐇 𝐛𝐫 and 𝐏𝐀𝐓𝐇 𝐛𝐮𝐬 have been determined for 𝑏𝑟𝑖 ∈ Ω𝐿𝐼𝑁 . Then,
𝑓𝑎𝑖𝑙 𝑓𝑎𝑖𝑙 𝑓𝑎𝑖𝑙
𝑭𝑨𝒃𝒖𝒔 {𝑏𝑟𝑖 } is obtained searching 𝐏𝐀𝐓𝐇 𝐛𝐫 {𝑏𝑟𝑖 } and 𝐏𝐀𝐓𝐇 𝐛𝐫 {𝑏𝑟𝑖 } using the proposed
algorithm described in Table C. 1 in Appendix C.
For the case with multiple faults, a Hadamard multiplication ∏ is used according to 𝐅𝐀𝐛𝐮𝐬 {𝒇𝒍𝒐𝒄𝑡 }.
Eq. (5.108) shows the formulation for determining 𝑩𝑭𝒂𝑡 based on 𝒇𝒍𝒐𝒄𝑡 .
𝑩𝑭𝒂𝑡 = ∏𝑖∈Ω𝑓𝑙𝑜𝑐[∘ 𝐅𝐀𝐛𝐮𝐬 (𝑓𝑙𝑜𝑐𝑖 , : )] (5.108)
Similarly, 𝑩𝑭𝒂𝑡 is repeatedly determined for 𝑡 ∈ Ω𝑇 and then stored in 𝐁𝐅𝐚. For the case where there
is no fault at time 𝑡, the methodology described above can be skipped and 𝑩𝑭𝒂𝑡 can be simply set
vector containing the element ‘1’. More detail elaborates the determination of 𝑩𝑭𝒂𝑡 until 𝐁𝐅𝐚 can
be found in Appendix C
5.5.2.3 Microgrid Formation Set
The matrix 𝐌𝐆𝑡𝐬𝐞𝐭 is used for indicating the set of possible microgrid formations once the fault is
eliminated. In this sense, taking advantage of the Depth First Search (DFS) technique the matrix
𝐌𝐆𝑡𝐬𝐞𝐭 for a generic 𝒇𝒍𝒐𝒄𝑡 and 𝑰𝑺𝑾𝑠𝑖𝑡𝑒 is obtained employing the proposed algorithm described in
Table C. 2 in Appendix C for all 𝑡 ∈ Ω𝑇 . This is possible as the DFS technique allows the proposed
algorithm able to track a set of microgrid(s) according to their roots noted by 𝑟𝑜𝑜𝑡 𝑀𝐺 which is
corresponding to the location of the switch(es). More details about the DFS can be found in [109],
[110]. The mathematically example detailed the 𝐌𝐆𝑡𝐬𝐞𝐭 can be found in Appendix C.
5.5.2.4 Microgrid Formation Indicators
In this section, the time dependent microgrid formation indicators are determined. These indicators
are the binary matrices and binary vectors representing the time-coupled microgrid tropology and
operation especially during the N-K contingencies scenarios. The methodology applied for
determining these indicators are described in the following subsection.
109
a) System Buses Operation Mode
The system buses operation mode are described by matrix 𝐛𝐦 . Matrix 𝐛𝐦 assembles the binary
vectors 𝒃𝒎
𝑡 containing the binary elements responsible for indicating the operation mode, i.e., GC or
IS, of each system bus; where, the element of 𝒃𝒎

𝑡 is “1” if the corresponding bus is located at
microgrid area, otherwise “0” for those buses connected to the main grid. The formation of 𝒃𝒎
𝑡 can
be expressed by (5.126). For the determine 𝒃𝒎

𝑡 , the algorithm in Table C. 3 in Appendix C is adopted.
𝒃𝒎 𝑚 𝑚 𝑚
𝑡 = [𝑏1,𝑡 , 𝑏2,𝑡 , … , 𝑏𝑛𝑏𝑢𝑠 ,𝑡 ] (5.109)
b) Non-Energized System Buses
The microgrid can be operated in IS mode when there is at least a single DDG unit localized within
the microgrid. In this sense, let a vector 𝑴𝑮𝒔𝒆𝒕
𝑚,𝑡 define the set of 𝑚
𝑡ℎ
microgrid found at time 𝑡, if
there is no DDG unit installed in 𝑴𝑮𝒔𝒆𝒕 𝒔𝒆𝒕
𝑚,𝑡 , this 𝑴𝑮𝑚,𝑡 can not operate in IS mode. For this reason, a
binary vector 𝒃𝒖𝒔𝑛𝑒

𝑡 is adopt for indicating that the buses within the microgrid cannot be energized
due to the lack of local DDG(s); where, the element of 𝒃𝒖𝒔𝑛𝑒

𝑡 is “1” when the corresponding bus is
located in a with DDG(s), otherwise “0” for those buses located in microgrids lacking local DDG(s).
The general formation for 𝒃𝒖𝒔𝑛𝑒
𝑡 is shown in (5.110). Then, the general algorithm for determiner
𝑡 for an instance time 𝑡 ∈ Ω 𝑇 is shown in Table C. 4 of Appendix C. See Appendix C for more
detail exemplifies the formation of 𝒃𝒖𝒔𝑛𝑒

𝑡 .
𝒃𝒖𝒔𝑛𝑒 𝑛𝑒 𝑛𝑒 𝑛𝑒
𝑡 = [𝑏𝑢𝑠1,𝑡 , 𝑏𝑢𝑠2,𝑡 , … , 𝑏𝑢𝑠𝑛𝑏𝑢𝑠 ,𝑡 ] (5.110)
c) Frequency-Controlled Buses and Deactivated DDGs
According to (5.70) there is a need to identify the location of buses with frequency control capability
within the microgrid. Nonetheless, if more than a single microgrid is created, the frequency-controlled
bus for each formed microgrid should be identified. For this, the vector 𝒇𝒃𝒖𝒔
𝑡 is employed. Its elements
indicate the location of frequency-controlled buses for each DDG. Still, there might be cases where a
DDG is located within the fault area. In this case, the binary vector 𝑫𝑫𝑮𝑑𝑜𝑤𝑛
𝑡 is used, where elements
corresponding to DDGs standing in fault areas are set to zero, otherwise set to “1”. (5.111) and (5.112)
shows the formation for 𝒇𝒃𝒖𝒔
𝑡 and 𝑫𝑫𝑮𝑑𝑜𝑤𝑛
𝑡 , respectively. The algorithm adopted for determining
𝒇𝒃𝒖𝒔
𝑡 is show in Table C. 5 in Appendix C.
𝒇𝒃𝒖𝒔
𝑡
𝑏𝑢𝑠 𝑏𝑢𝑠
= [𝑓1,𝑡 , 𝑓2,𝑡 , … , 𝑓𝑛𝑏𝑢𝑠
𝑑𝑑𝑔 ,𝑡
] (5.111)
110
𝑫𝑫𝑮𝑑𝑜𝑤𝑛
𝑡
𝑑𝑜𝑤𝑛
= [𝐷𝐷𝐺1,𝑡 𝑑𝑜𝑤𝑛
, 𝐷𝐷𝐺2,𝑡 , … , 𝐷𝐷𝐺𝑛𝑑𝑜𝑤𝑛
𝑑𝑑𝑔 ,𝑡
] (5.112)
d) WT, PV and ESS Operation Status
WT, PV and ESS cannot maintain the microgrid operation in IS without DDG support. In this sense,
WTs, PVs and ESSs that are located inside non-successful IS microgrid should be deactivated. For
this, vectors indicating the WTs, PVs and ESSs deactivation status can be determined by simply
mapping their location with 𝒃𝒖𝒔𝑛𝑒
𝑡 and can be as expressed by (5.113)-(5.115).
𝑾𝑻𝑑𝑜𝑤𝑛
𝑡 = 𝒃𝒖𝒔𝑛𝑒 𝑠𝑖𝑡𝑒
𝑡 (𝑾𝑻𝑠𝑝 ) (5.113)
𝑷𝑽𝑑𝑜𝑤𝑛
𝑡 (𝑷𝑽𝑠𝑝 ) (5.114)
𝑬𝑺𝑺𝑑𝑜𝑤𝑛
𝑡 (𝑬𝑺𝑺𝑠𝑝 ) (5.115)
e) Time-Coupled 𝒀𝒃𝒖𝒔
The initial system impedance is determined via the original system topology, i.e., 𝑍𝑖,𝑡,𝑠,𝑠𝑒 = 𝑍𝑖0 . Still,
during severe event analysis the impedance of overhead line branches changes as the system topology
adapts, i.e., opening the switch of an overhead line leads to an equivalent impedance 𝑍𝑖,𝑡,𝑠,𝑠𝑒 = ∞. In
this sense, the binary matrix 𝐋𝐅𝐚 is used for updating 𝑍𝑖,𝑡,𝑠,𝑠𝑒 at an instance time 𝑡, where the element
of 𝐋𝐅𝐚 can be used as a multiplication for modifying the impedance 𝑍𝑖,𝑡,𝑠,𝑠𝑒 as expressed in (5.116).
𝑍𝑖,𝑡,𝑠,𝑠𝑒 = (1/𝐿𝐹𝑎𝑖,𝑡,𝑠,𝑠𝑒 )𝑍𝑖0 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.116)
𝐛𝐮𝐬
The time-coupled 𝐘𝑠,𝑠𝑒 can be expressed by (5.117) using the methodology found in [18].
Conductance matrix 𝐆 and susceptance matrix 𝐁 along the planning horizon Ω𝑇 can be obtained using
(5.118) and (5.119).
𝐛𝐮𝐬 𝐛𝐮𝐬
𝐛𝐮𝐬
𝐘𝑠,𝑠𝑒 = [𝐘1,𝑠,𝑠𝑒 , 𝐘2,𝑠,𝑠𝑒 , … , 𝐘𝑁𝐛𝐮𝐬
𝑇 ,𝑠,𝑠𝑒
] ,∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.117)
𝐛𝐮𝐬
𝐆𝑠,𝑠𝑒 = 𝑟𝑒𝑎𝑙(𝐘𝑠,𝑠𝑒 ) (5.118)
𝐛𝐮𝐬
𝐁𝑠,𝑠𝑒 = 𝑖𝑚𝑎𝑔(𝐘𝑠,𝑠𝑒 ) (5.119)
111
5.5.3 Integration of Microgrid Formation Indicators to Inner Level Optimization
In this section, the microgrid formation indicators are integrated in the LOPF problem allowing the
capture of multi-microgrid topology and the status of available resources over the planning horizon
Ω𝑇 . For this, the LOPF formulation updated according to the multi-microgrid formation indicators
are listed as follows:
a) Objective Function
With a new 𝒃𝒎
𝒕 , the formulation (5.58) and (5.62) were replaced by (5.120) and (5.121) for introduce
𝑚
𝑏𝑖,𝑡,𝑠,𝑠𝑒 :
𝑃𝐶𝐶 𝑚
𝐶𝑡,𝑠,𝑠𝑒 (1 − 𝑏𝑖=1,𝑡,𝑠,𝑠𝑒 )
𝐷𝐷𝐺 𝑚
𝑀𝐼𝐿𝑃,𝑀𝐺𝑠
+𝐶𝑡,𝑠,𝑠𝑒 (1 − 𝑏∀𝑖∈𝛺𝐷𝐷𝐺 ,𝑡,𝑠,𝑠𝑒
)
𝑂𝐹𝑠,𝑠𝑒 = ∑𝑡∈𝛺𝑇 𝐶𝑈𝑅𝑇 𝑚 ,∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.120)
+𝐶𝑡,𝑠,𝑠𝑒 𝑏∀𝑖∈𝛺𝐵𝑈𝑆,𝑡,𝑠,𝑠𝑒
𝑉𝐼,𝑀𝐺𝑠
[ +𝐶𝑡,𝑠,𝑠𝑒 ]
𝑚
𝑉𝐼,𝑀𝐺𝑠 𝑛𝑒 (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒 )𝑉 𝐺𝐶,𝑠𝑝𝑒𝑐
𝐶𝑡,𝑠,𝑠𝑒 = 𝑉𝑏𝑎𝑠𝑒 (𝑃𝐶 𝑉𝐼 [∑𝑖∈𝛺𝐵𝑈𝑆 𝑏𝑢𝑠𝑖,𝑡,𝑠,𝑠𝑒 ( 𝑚 )]) ∆𝑡
+𝑏𝑖,𝑡,𝑠,𝑠𝑒 𝑉𝑖∗∗ − 𝑉𝑖,𝑡,𝑠,𝑠𝑒
,∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.121)
b) Constraints
The update constraints considering the microgrids formulation indicators can be listed as follows:
(1 − 𝑏𝑢𝑠𝑛𝑒 𝑐𝑢𝑟𝑡 𝑚
𝑖,𝑡,𝑠,𝑠𝑒 ) ≤ 𝑏𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑏𝑖,𝑡,𝑠,𝑠𝑒
,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.122)
𝑃𝑖,𝑡,𝑠,𝑠𝑒 + 𝑃𝑖,𝑡,𝑠,𝑠𝑒
∑𝑗∈Ω𝐵𝑈𝑆 𝐺𝑖𝑗,𝑡,𝑠,𝑠𝑒 𝑉𝑖,𝑡,𝑠,𝑠𝑒 𝑊𝑇 𝑃𝑉
+𝑃𝑖,𝑡,𝑠,𝑠𝑒 + 𝑃𝑖,𝑡,𝑠,𝑠𝑒
𝑃𝑖,𝑡,𝑠,𝑠𝑒 =( )= 𝑑𝑐ℎ
− ∑𝑗∈Ω𝐵𝑈𝑆 𝐵𝑖𝑗,𝑡,𝑠,𝑠𝑒 𝛿𝑗,𝑡,𝑠,𝑠𝑒 +𝑃𝑖,𝑡,𝑠,𝑠𝑒 − 𝜂𝑖𝐸𝑆𝑆 𝑃𝑖,𝑡,𝑠,𝑠𝑒
𝑐ℎ
( −𝑃𝑖,𝑡,𝑠,𝑠𝑒 (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒 ) )
,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.123)
+𝑄𝑖,𝑡,𝑠,𝑠𝑒 + 𝑄𝑖,𝑡,𝑠,𝑠𝑒
− ∑𝑗∈Ω𝐵𝑈𝑆 𝐵𝑖𝑗,𝑡,𝑠,𝑠𝑒 𝑉𝑖,𝑡,𝑠,𝑠𝑒 𝐸𝑆𝑆
𝑄𝑖,𝑡,𝑠,𝑠𝑒 =( )=( +𝑄𝑖,𝑡,𝑠,𝑠𝑒 )
− ∑𝑗∈Ω𝐵𝑈𝑆 𝐺𝑖𝑗,𝑡,𝑠,𝑠𝑒 𝛿𝑗,𝑡,𝑠,𝑠𝑒
−𝑄𝑖,𝑡,𝑠,𝑠𝑒 (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒 )
,∀𝑖 ∈ Ω𝐵𝑈𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.124)
112
𝑏𝑟
𝑔𝑖(𝑘)𝑗(𝑘),𝑡,𝑠,𝑠𝑒 (𝑉𝑖,𝑡,𝑠,𝑠𝑒 − 𝑉𝑗,𝑡,𝑠,𝑠𝑒 )
𝑃𝑘,𝑡,𝑠,𝑠𝑒 =( )
+𝑏𝑖(𝑘)𝑗(𝑘),𝑡,𝑠,𝑠𝑒 (𝛿𝑖,𝑡,𝑠,𝑠𝑒 − 𝛿𝑗,𝑡,𝑠,𝑠𝑒 )
,∀𝑘 ∈ Ω𝐿𝐼𝑁 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.125)
𝑏𝑟
𝑏𝑖(𝑘)𝑗(𝑘),𝑡,𝑠,𝑠𝑒 (𝑉𝑖,𝑡,𝑠,𝑠𝑒 − 𝑉𝑗,𝑡,𝑠,𝑠𝑒 )
𝑄𝑘,𝑡,𝑠,𝑠𝑒 =( )
+𝑔𝑖(𝑘)𝑗(𝑘),𝑡,𝑠,𝑠𝑒 (𝛿𝑖,𝑡,𝑠,𝑠𝑒 − 𝛿𝑗,𝑡,𝑠,𝑠𝑒 )
,∀𝑘 ∈ Ω𝐿𝐼𝑁 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.126)
𝑝
𝐷𝐷𝐺
[− 𝑘 𝑖𝜔 (𝛿𝑓𝑀𝐺
𝑏𝑢𝑠
,𝑡,𝑠,𝑠𝑒
𝑚
− 𝜋⁄2)⁄𝑚𝑖 ] 𝑏𝑖,𝑡,𝑠,𝑠𝑒
𝑃𝑖,𝑡,𝑠,𝑠𝑒 =( 𝑖,𝑡 )
𝑚 𝐷𝐷𝐺
+(1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒 )𝑃𝑖,𝑡,𝑠,𝑠𝑒
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.127)
𝐷𝐷𝐺 𝑞
𝑄𝑖,𝑡,𝑠,𝑠𝑒 = [(𝑉𝑖∗ − 𝑉𝑖,𝑡,𝑠,𝑠𝑒 + 𝑢𝑖,𝑡,𝑠,𝑠𝑒
𝑣 𝑚
)⁄𝑛𝑖 ]𝑏𝑖,𝑡,𝑠,𝑠𝑒 𝑚
+ (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒 𝐷𝐷𝐺
)𝑄𝑖,𝑡,𝑠,𝑠𝑒
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.128)
𝑚
𝑉𝑖,𝑡,𝑠,𝑠𝑒 𝑏𝑖,𝑡,𝑠,𝑠𝑒 = 𝑉𝑖∗∗ 𝑏𝑖,𝑡,𝑠,𝑠𝑒
𝑚
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.129)
𝐷𝐷𝐺
0 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 𝐷𝐷𝐺𝑖,𝑡,𝑠,𝑠𝑒
𝑑𝑜𝑤𝑛
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.130)
𝜒𝑖𝐷𝐷𝐺,𝑚𝑖𝑛 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 𝐷𝐷𝐺𝑖,𝑡,𝑠,𝑠𝑒

𝑑𝑜𝑤𝑛 𝐷𝐷𝐺
≤ 𝑄𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝜒𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 𝐷𝐷𝐺𝑖,𝑡,𝑠,𝑠𝑒
𝑑𝑜𝑤𝑛
,∀𝑖 ∈ Ω𝐷𝐷𝐺 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.131)
𝑊𝑇,𝑢𝑠𝑒 𝑊𝑇 𝑑𝑜𝑤𝑛
0 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝑊𝑇𝑖,𝑡,𝑠,𝑠𝑒 ,∀𝑖 ∈ Ω𝑊𝑇 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.132)
𝑃𝑉,𝑢𝑠𝑒 𝑃𝑉 𝑑𝑜𝑤𝑛
0 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 𝑃𝑉𝑖,𝑡,𝑠𝑒 ,∀𝑖 ∈ Ω𝑃𝑉 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.133)
𝑑𝑐ℎ
0 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 𝑏𝐸𝑆𝑆 𝑑𝑜𝑤𝑛
𝑖,𝑡,𝑠,𝑠𝑒 𝐸𝑆𝑆𝑖,𝑡,𝑠,𝑠𝑒
,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.134)
𝑐ℎ
0 ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 (1 − 𝑏𝐸𝑆𝑆 𝑑𝑜𝑤𝑛
𝑖,𝑡,𝑠,𝑠𝑒 )𝐸𝑆𝑆𝑖,𝑡,𝑠,𝑠𝑒
,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.135)
𝜒𝑖𝐸𝑆𝑆,𝑚𝑖𝑛 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 𝐸𝑆𝑆𝑖,𝑡,𝑠,𝑠𝑒

𝑑𝑜𝑤𝑛 𝐸𝑆𝑆
≤ 𝑄𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝜒𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 𝑆𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 𝐸𝑆𝑆𝑖,𝑡,𝑠,𝑠𝑒
𝑑𝑜𝑤𝑛
,∀𝑖 ∈ Ω𝐸𝑆𝑆 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′

𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.136)
113
−𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒
𝑚 𝑃𝐶𝐶
) ≤ 𝑃𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒
𝑚
)
,∀𝑖 ∈ 𝛺𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.137)
𝜒𝑖𝑇𝑅,𝑚𝑖𝑛 𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒

𝑚 𝑃𝐶𝐶
) ≤ 𝑄𝑖,𝑡,𝑠,𝑠𝑒 ≤ 𝜒𝑖𝑇𝑅,𝑚𝑎𝑥 𝑆𝑖𝑇𝑅,𝑚𝑎𝑥 (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒
𝑚
)
,∀𝑖 ∈ 𝛺𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.138)
𝑚 𝑚
𝑉𝑖,𝑡,𝑠,𝑠𝑒 (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒 ) = 𝑉 𝐺𝐶,𝑠𝑝𝑒𝑐 (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒 )
,∀𝑖 ∈ 𝛺𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.139)
𝑃𝐶𝐶 𝑚
𝛿𝑖,𝑡,𝑠,𝑠𝑒 (1 − 𝑏𝑖,𝑡,𝑠,𝑠𝑒 )=0 ,∀𝑖 ∈ 𝛺𝑃𝐶𝐶 , ∀𝑡 ∈ 𝛺𝑇 , ∀𝑠 ∈ Ω′𝑠 ∪ Ω′′
𝑠 , ∀𝑠𝑒 ∈ Ω𝑠𝑒 (5.140)
First, (5.64) is replaced by (5.122) for allowing the load at the individual bus curtail during IS mode,
while forcing 𝑏𝑐𝑢𝑟𝑡
𝑖,𝑡 = 1 for any bus 𝑖 that corresponding to the bus non-energize 𝑏𝑢𝑠𝑛𝑒
𝑖,𝑡,𝑠,𝑠𝑒 . Then,
equations (5.123)-(5.126) are power flow constraints considering the microgrids topology using the
𝐛𝐮𝐬
time-coupled 𝐘𝑠,𝑠𝑒 . The updated constraints for the DDG units are expressed by (5.127)-(5.128).
𝑚
Comparing to (5.70)-(5.72), 𝑏𝑖,𝑡,𝑠,𝑠𝑒 is introduced in (5.127)-(5.129), allowing the problem
consideration individual DDG units operation, i.e., GC or IS mode. Moreover, for (5.127), 𝛿𝑓𝑀𝐺
𝑏𝑢𝑠
,𝑡,𝑠,𝑠𝑒
𝑖,𝑡
is dependent to frequency-controlled buses. Eqs. (5.77) and (5.78) are replaced by (5.130) and (5.131)
for accounting the status of the DDG units i.e., online or failure. The inequalities (5.132)-(5.136)
guarantee that WTs, PVs and ESSs are deactivated in case they are located inside a non-successful IS
microgrids. Finally, eqs. (5.137)-(5.140) defined the updated power exchange constraints by
𝑚
introducing 𝑏𝑖,𝑡 .
c) MILP Model for LOPF Considering Microgrids Formation
According to the updated LOPF problem formulation the MILP model used for the deep-learning
approach can be summarized as following:
𝑀𝐼𝐿𝑃,𝑀𝐺𝑠
min 𝑂𝐹𝑠,𝑠𝑒 (5.141)
𝑿𝑀𝐼𝐿𝑃
Subject to:
• Voltage and load curtailment constraints: (5.63), (5.122)
• Power flow constraints: (5.123)-(5.126), (5.69)
114
• DDG operation constrains: (5.73)-(5.75), (5.78), (5.79), (5.127)-(5.131)
• WT and PV operation constrains: (5.133), (5.134)
• ESS operation constraints: (5.82), (5.86)-(5.89), (5.134)-(5.136)
• Power exchange constraints: (5.92), (5.93), (5.137)-(5.140)
5.5.4 Deep Neural Network (DNN) Model
At this stage, for each sample 𝑠𝑝, the identified targets (5.102)-(5.104) can be calculated using the
results obtained by the LOPF solution. Then, a set of sample inputs 𝐈𝐍 𝐝𝐧𝐧 and the calculated targets
(5.102)-(5.104) are used for training the DDN models, wherein 70% of the samples were arbitrarily
selected for the training procedure and the remaining samples for the testing procedure. For
simplicity, a single DNN model is trained for predicting the single identification target. In this work,
each DDN model employed is formed as a multi-layer feedforward neural network with three levels,
i.e., a single input layer, several hidden layers and a single output layer [111][112]. Figure 5. 9 shows
the basic scheme for a multi-layer feed-forward neural network architecture.
ninr×nfeat ntar nnnrhl

a0dnn aidnn
Wi W(nₕₗ+1)
nfeat×1 ninr×1
pidnn pdnn'
fidnn f dnn'
ninr×1 ntar×1
1 bi 1 b(nₕₗ+1) ntar×1
nfeat ninr×nfeat ntar×1 adnn'
INPUT HIDDEN LAYERS OUTPUT LAYER
FIGURE 5. 9: General scheme of multi-layer feed-forward neural network architecture for

predicting a single identified target
According 𝐈𝐍 𝐝𝐧𝐧 and targets (5.102)-(5.104), a general DNN model architecture designed for a single
sample can be expressed by (5.142)-(5.144).
𝒂𝒅𝒏𝒏 𝒅𝒏𝒏
0,𝑠𝑝 = 𝑰𝑵𝑠𝑝 ,𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.142)
𝒂𝒅𝒏𝒏
𝑖,𝑠𝑝 = 𝑓𝑖
𝑑𝑛𝑛
(𝐖𝑖 𝒂𝒅𝒏𝒏
𝑖−1,𝑠𝑝 + 𝐛𝑖 ) ,𝑖 = 1,2, . . , 𝑛ℎ𝑙 , 𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.143)
𝑑𝑛𝑛 ′ ′
𝑎𝑠𝑝 = 𝑓 𝑑𝑛𝑛 (𝐖𝑛ℎ𝑙 +1 𝒂𝒅𝒏𝒏
𝑛ℎ𝑙 ,𝑠𝑝 + 𝐛𝑛ℎ𝑙 +1 ) ,𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.144)
115
The input layer of the hidden layer receives the vector of input 𝒂𝒅𝒏𝒏
0,𝑠𝑝 expressed by (5.142). Then, each
of layer in the hidden layer compute their neuron pre-activation 𝒑𝒅𝒏𝒏

𝑖,𝑠𝑝 regarding to the weight matrix
and biases, the activation function 𝑓𝑖𝑑𝑛𝑛 (. ) employed to 𝒑𝒅𝒏𝒏

𝑖,𝑠𝑝 for delivering the neuron activation for
the next layer (5.143). At the end, the output layer computes a single prediction neuron activation,
𝑛𝑡𝑎𝑟 = 1, using the neuron activation 𝒂𝒅𝒏𝒏
𝑛ℎ𝑙 ,𝑠𝑝 that is obtained by the last layer of the hidden layer
(5.144). In this work, 𝑓𝑖𝑑𝑛𝑛 (. ) is a sigmoid activation function for the hidden layer, while the output
′
layer linear activation function denoted by 𝑓 𝑑𝑛𝑛 (. ) is used. Finally, the DNN model has been trained
via the Bayesian regularization backpropagation (BRB) algorithm [111]. This algorithm is a
backpropagation-based approach focused on minimizing the mean squares sum of the network errors
(MSE) during the DDN model training. Moreover, seeking to minimized and make the network
response smoother, as well as likely reducing overfitting impact [111], [113]. The proposed algorithm
also includes the network regularization based on the network sum of mean square weights (MSW).
As can be seen in Figure 5. 9, the prediction obtained by the trained DDN model should satisfy the
criteria. In this work, the errors indicated in (5.145) have been used as a criteria calculated for every
round of DNN training completion. Then, (5.144) and (5.145) are calculated for illustrating the
performance of the model for different types of errors.
|𝑇𝐺𝑠𝑝 −𝑛𝑒𝑡 𝑛 (𝑰𝑵𝒅𝒏𝒏

𝑠𝑝 )| 𝑡𝑒𝑠𝑡
𝑒1𝑑𝑛𝑛 = 100% × max ( 𝑇𝐺𝑠𝑝
) ,∀𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.145)
𝑡𝑒𝑠𝑡
𝑒2𝑑𝑛𝑛 = mean(|𝑇𝐺𝑠𝑝 − 𝑛𝑒𝑡 𝑛 (𝑰𝑵𝒅𝒏𝒏
𝑠𝑝 )|) ,∀𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.144)
𝑚𝑒𝑎𝑛(|𝑇𝐺𝑠𝑝 −𝑛𝑒𝑡 𝑛 (𝑰𝑵𝒅𝒏𝒏

𝑠𝑝 )|) 𝑡𝑒𝑠𝑡
𝑒3𝑑𝑛𝑛 = 100% × ,∀𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.145)
𝑚𝑒𝑎𝑛(𝑇𝐺𝑠𝑝 )
Where, 𝑛𝑒𝑡 𝑛 (𝑰𝑵𝒅𝒏𝒏 𝒅𝒏𝒏

𝑠𝑝 ) is the prediction results for the input sample 𝑰𝑵𝑠𝑝 using DNN model
corresponding to target 𝑛. First, 𝑒1𝑑𝑛𝑛 represents the relative error between the prediction and real
value. In this case, we apply a maximum for delivering the worst error over the vector of relative
error. Then, 𝑒2𝑑𝑛𝑛 and 𝑒3𝑑𝑛𝑛 represent the mean absolute and relative errors, presenting more
generalized perspectives in comparison to 𝑒1𝑑𝑛𝑛 .
In this work, the hyper parameters are randomly selected for every round of DNN training, i.e., (a)
activation function for hidden layers, (b) activation function for outer layer, (c) number of layers and
corresponding neurons in the hidden layers. According to the BRB algorithm, the training round stops
if the sum squared error and the sum squared weights are not improved over subsequent training
116
iterations (epochs) [111], [113]. Then, trained DNN models are validated though the testing set until
satisfying criteria 𝑒1𝑑𝑛𝑛 threshold, guaranteeing that the worst relative error between the target and
prediction is less than 𝑒1𝑑𝑛𝑛 . Finally, trained DNN models are used for predicting the targets and
calculating the LOPF dependent indices 𝑅𝐸𝑆𝐼 𝑂𝐿𝐸𝑉 , 𝐴𝑂𝑀𝐶 𝐷𝐷𝐺 , 𝐴𝐸𝑀𝐼𝑆 𝐷𝐷𝐺 and 𝐴𝐸𝑀𝐼𝑆 𝐺𝑅𝐼𝐷 for the
outer level. In this sense, let 𝑛𝑒𝑡 𝑛 (𝑿𝑶𝑳𝑬𝑽 ) be a trained DDN network for target 𝑛 depending on
𝑿𝑶𝑳𝑬𝑽, the LOPF dependent indices for the outer level calculated using the trained DNN model can
be expressed by (5.146)-(5.152).
𝑅𝐸𝑆𝐼 𝑂𝐿𝐸𝑉,𝑑𝑙 = 𝑛𝑒𝑡1 (𝑿𝑶𝑳𝑬𝑽 ) ,∀𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.146)
𝑇𝐶 𝑂𝐿𝐸𝑉,𝑑𝑙 = 𝐶𝐷𝐷𝐺,𝑑𝑙 + 𝐶 𝑊𝑇 + 𝐶 𝑃𝑉 + 𝐶 𝐸𝑆𝑆 + 𝐶 𝐼𝑆𝑊 (5.147)
𝐶𝐷𝐷𝐺,𝑑𝑙 = 𝐶𝑃𝐶 𝐷𝐷𝐺 + 𝑇𝐿 × 𝐴𝑂𝑀𝐶𝐷𝐷𝐺,𝑑𝑙 (5.148)
365 𝜎 𝑓𝑢𝑒𝑙 𝐷𝐷𝐺

𝐴𝑂𝑀𝐶𝐷𝐷𝐺,𝑑𝑙 = 𝑛𝑒𝑡 2 (𝑿𝑶𝑳𝑬𝑽 ) ( 𝑆𝑏𝑎𝑠𝑒 ( + 𝜎𝑂&𝑀 )) (5.149)
4 𝜂𝐷𝐷𝐺
𝐴𝐸𝑀𝐼𝑆 𝑂𝐿𝐸𝑉,𝑑𝑙 = 𝐴𝐸𝑀𝐼𝑆𝐷𝐷𝐺,𝑑𝑙 + 𝐴𝐸𝑀𝐼𝑆 𝐺𝑅𝐼𝐷,𝑑𝑙 (5.150)
365
𝐴𝐸𝑀𝐼𝑆𝐷𝐷𝐺,𝑑𝑙 = 𝑛𝑒𝑡 2 (𝑿𝑶𝑳𝑬𝑽 )( 𝑆 𝐾 𝐷𝐷𝐺 )
,∀𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.151)
365
𝐴𝐸𝑀𝐼𝑆 𝐺𝑅𝐼𝐷,𝑑𝑙 = 𝑛𝑒𝑡 3 (𝑿𝑶𝑳𝑬𝑽 )( 𝑆 𝐾 𝐺𝑅𝐼𝐷 ) ,∀𝑠𝑝 ∈ Ω𝑆𝐴𝑀 (5.152)
5.6 Stage IV: Solution for Optimum Planning

The proposed multi-objective optimization problem incorporating the DNN models for the resilience-
oriented planning of microgrids can be summarized as follows:
min (𝑅𝐸𝑆𝐼 𝑂𝐿𝐸𝑉,𝑑𝑙 , 𝑇𝐶 𝑂𝐿𝐸𝑉,𝑑𝑙 , 𝐴𝐸𝑀𝐼𝑆 𝑂𝐿𝐸𝑉,𝑑𝑙 ) (5.153)

𝑿𝑶𝑳𝑬𝑽
Subject to:
• DERs size constraint: (5.47)-(5.51)
• Maximum allowable bus power constraint: (5.52)
• ISWs and DERs installation constraints: (5.53)-(5.57)
The proposed problem has been solved using the population-based multi-objective optimization
approach NSGA-II [114]. Figure 5. 10 shows the general outline of the proposed optimum planning
NSGA-II based on DNN models.
117
START
Outer level: NSGA-II

Generation counter gen=0
yes gen = 0
Generate npop
random initial no
parent population
Generate child npop offspring

population using selection,
crossover and mutation
Generation counter ind=0
Calcul OF functions
Determine XOLEV directly
dependent indices
Inner level: DL-LOPF
Determine:
RESIolev,dl, AOMCDDG,dl,
AEMISDDG,dl and AEMISGRID,dl
B A
Calculate multi-objective function by:

Import RESIolev,dl from A and determine
TColev and AEMISolev from A&B
ind=ind+1 ind >npop

yes gen=gen+1
gen > 0
yes
Mixed parent population

with child population
Non-dominated sorting the

set of 2npop mixed population
Form new npop parant propulation
satisfy
yes
Non-dominated solutions of the

pareto front and the set of
solutions
END
FIGURE 5. 10: Optimum planning approach
118
The NSGA-II is widely used for multi-objective optimization problems due to its computational
efficient and faster convergence compared to the another techniques [114]. Moreover, the solution
delivered by NSGA-II is closed to the true Pareto-optima frontier [114]. Additional details on NSGA-
II and its implementation can be found in [114].
In the beginning, the first iteration at 𝑔𝑒𝑛 = 0 the NSGA-II starts generating an initial random parent
population with 𝑛𝑝𝑜𝑝 individuals. In this work, a single individual includes the combination of the
genes representing the allocation of the planning elements i.e., ISWs, DDGs, WTs, PVs and ESSs.
Figure 5. 11 shows the structure of an individual or a chromosome. Then, considering each single
individual, the multi-objective function can be determined from two directions A and B. For the
following iteration 𝑔𝑒𝑛 > 0 the NSGA-II starts forming an 𝑛𝑝𝑜𝑝 child offspring population using the
genetic operators, i.e., selection, crossover and mutation [114]. Similar to the parent population, the
multi-objective functions are evaluated for 𝑛𝑝𝑜𝑝 child individuals, and a new parent population is
determined.
 can
ISW  can
DDG WT
can
 can
PV  can
ESS
ISW1 ... ISWn DDG 1 ... DDG n WT1 ... WTn PV 1 ... PV n ESS1 ... ESSn
Binary genes Binary genes Binary genes Binary genes Binary genes
for ISW for DDG for WT for PV for ESS
Single Individual or Chomosome
FIGURE 5. 11: Structure of chromosome for NSGA-II
First, the child population is combined with the parent population forming a mixed population with
2 × 𝑛𝑝𝑜𝑝 individuals. Then, a new parent population is developed selecting 𝑛𝑝𝑜𝑝 individuals from
the mixed population using the fast non-dominated sorting technique [114]. According to this
technique, individual(s) classified in the lowest rank of the frontier are selected and moved up to the
highest frontier until the quantity of selected individuals reach 𝑛𝑝𝑜𝑝 . If the last ranked individuals
cause the quantity to exceed 𝑛𝑝𝑜𝑝 , the crowding distance is applied for selecting the remainder
individuals. This process is repeated until satisfy the NSGA-II criteria.
5.7 Numerical Results

In order to validate the proposed method, the IEEE 33-bus test system is selected. This system is a
typical distribution network with the ability to disconnected from the main grid via the main isolated
switch, i.e., ISW sub, installed at the upstream interfacing substation. Distribution lines infrastructure
119
are assumed as typical overhead power line, with loads weighing coefficient proportional to the
annual peak load, Figure 5. 12, similar to the work found in [3].
FIGURE 5. 12: Seasonal annual load pattern
The wind speed and irradiance data sets are obtained via the scenario generation procedure describe
in Section 5.3 using fast forward reduction. For this, data series extracted from NREL [115]
comprising the hourly profile for 10 years, respectively 2008-2017 for Victoria-BC, Canada, of wind
speed in MPH (miles per hours) and for irradiance in w/m2 are employed. Details pertaining to the
obtained set of scenarios reduction are shown in Table G. 6 in the Appendix G. Based on the obtained
data sets, the most vulnerable wind scenarios summarized in Table G. 7 are used for generating severe
𝑝𝑟𝑒
operation scenarios. For accounting the uncertainty of the N-K contingencies, 𝑁𝑠 is set sufficiently
𝑝𝑟𝑒
high, 𝑁𝑠 = 10000, allowing for the selection of 5 scenarios for each season 𝑠𝑒. This procedure
steps are detailed depicted in Section 5.3, and the parameters related to the scenario generation are
summarized in the Table G. 8. The daily market electricity price is extracted from NYISO [89], where
corresponding days and years are shown in Table G. 6 and Table G. 7 in the Appendix G. The test
system candidate buses and branches for the planning elements are shown in Figure 5. 13, with main
characteristics summarized in Table G. 9 in the Appendix G [1].
23 24 26 27 28 29 30 31 32
23 24 26 27 28 29 30 31 32
25 33
22 25
ISWsub 1 2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
MG
PCC 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
18
19 20 21
Candidate branch Microgrid PCC
19 20 21 22 Candidate bus Substation
Natural gas Wind turbine Photovoltaic Energy storage system Automatic recloser
Planning Elements
FIGURE 5. 13: 33-bus test system with candidate buses and branches
120
The proposed planning framework is developed in MATLAB environment, including the DNN
models training procedure and the implementation of NSGA-II. It is worth noting that the NaS
technology was selected for the ESSs due to its high number of lifecycles. In this work, the ESS
capacity rating is proportional to their power rating, similarly to the work in [116]. In this case, the
ratio between the rated capacity and rated power is set to 5 for all ESSs. This number was extracted
from the optimal solution found in [83]. Additionally, the maximum DoD is arbitrary set to 70% to
provide a balance between the battery lifecycle and its generation capability. The DDGs is the natural
gas type which can be operated in PQ mode, during GC mode, and droop/2nd, during IS mode. The
general parameters used in this research are summarized in Table G. 10 in the Appendix G.
5.7.1 Deep Learning Solution Validation
In this section, the performance of the trained DDN models is evaluated. As discussed in Section 5.5,
a single model is used for predicting the single target. The corresponding hyper-parameters for each
DNN model architecture are summarized in the Table 5. 3.
TABLE 5. 3: Hyper-parameters setting for DNN models’ architectures

Hyper-parameters DNN Tanning Targets
𝑻𝑮𝒔𝒑,𝟏 𝑻𝑮𝒔𝒑,𝟐 𝑻𝑮𝒔𝒑,𝟑
No. of hidden layer 3 3 3
No. of neurons for hidden layer 25-130-25 25-80-25 25-60-25
Activation fcn. for hidden layer log-sigmoid log-sigmoid log-sigmoid
Activation fcn. for output layer linear linear linear
log-sigmoid: Logistic sigmoid activation function
The number of samples are 40000 including 70% for the training set and 30% for testing set. Taking
advantage of BRB algorithm, training is terminated if the sum squared error and the sum squared
weights are not relatively improved over successive training iterations (epochs) [111], [113].
Additionally, an 𝑒1𝑑𝑛𝑛 threshold of 10% is considered for guaranteeing the accuracy of the DNN
models. In this case, the Figure 5. 14, Figure 5. 15 and Figure 5. 16 illustrate the training/testing
MSE and distributed error of the DNN model for predicting 𝑇𝐺𝑠𝑝,1 , 𝑇𝐺𝑠𝑝,1 and 𝑇𝐺𝑠𝑝,1 , respectively.
To this end, the performance of the trained DNN models are summarized in Table 5. 4.
121
model for predicting 𝑻𝑮𝟏
model for predicting 𝑻𝑮𝟐
122
model for predicting 𝑻𝑮𝟑
Analyzing Table 5. 4, one can draw conclusions regarding the DDN models errors for the three targets
of interest. First, the MSE, RMSE, and 𝑚𝑎𝑥|𝑦| for the training and testing procedures are found
around 10−4 to 10−7, 10−2 to 10−4, and 10−1 to 10−2 ranges, for all targets. These results indicate
the developed method significant ability to infer adequate optimal solutions, presenting errors
significantly smaller than accepted deviations. Where respective variations in the errors range are due
to the different units under consideration. In terms of the criteria, the 𝑒1𝑑𝑛𝑛 obtained for 𝑇𝐺𝑠𝑝,2 , 𝑇𝐺𝑠𝑝,2
and 𝑇𝐺𝑠𝑝,3 are found less than the 10% threshold. The 𝑒2𝑑𝑛𝑛 is found at 7.75 × 10−4, 1.06 × 10−2
kWh and 1.09 × 10−2 kWh for 𝑇𝐺𝑠𝑝,2 , 𝑇𝐺𝑠𝑝,2 and 𝑇𝐺𝑠𝑝,3 , respectively. Finally, 𝑒3𝑑𝑛𝑛 shows the ratio
in percentage for 𝑒2𝑑𝑛𝑛 over the mean target, leading to an error of less than 1%.
TABLE 5. 4: Performance of trained DNN models
Indicators Set Training targets
𝑇𝐺𝑠𝑝,1 𝑇𝐺𝑠𝑝,2 𝑇𝐺𝑠𝑝,3
- (kWh) (kWh)
-7
MSE Train 9.29×10 2.03×10 2.04×10-4
-4
Test 3.13×10-6 3.14×10-4 3.00×10-4

RMSE Train 9.64×10-4 1.43×10-2 1.43×10-2
Test 1.77×10-3 1.77×10-2 1.73×10-2
𝑚𝑎𝑥|𝑦| Train 1.50×10-2 1.36×10-1 1.23×10-1
Test 3.15×10-2 3.29×10-1 1.52×10-1
𝑒1𝑑𝑛𝑛 (%) - 7.96 1.97 0.09
𝑒2𝑑𝑛𝑛 (kWh) - 7.75×10-4 1.06×10-2 1.09×10-2
𝑒3𝑑𝑛𝑛 (%) - 1.44×10-1 2.80×10-2 0.63×10-2
123
5.7.2 NSGA-II Performance based DNN Model
In this section, the performance of the proposed NSGA-II based DNN model is evaluated. The
NSGA-II is responsible for generating the multi-objective problem solution according to the pareto
frontier. In this work, the maximum number of iterations and the population are set to 1000. The
crossover and mutation probability are set to 0.8 and 0.01, respectively. These parameters were
selected based on the guidelines available in [114], [117].
The solutions obtained including the 𝐗 𝐎𝐋𝐄𝐕 and the multi-objective functions values are based DNN
models’ prediction. In order to validate these solutions, first the LOPF problem (5.141) is re-simulated
for the obtained solution 𝐗 𝐎𝐋𝐄𝐕 . Then, the solution obtained by re-simulating the LOPF is used for
calculating the real values of multi-objective functions. Then, R-square (R2 ) and error indices (5.145)-
(5.145) are calculated for evaluating the error between real objective functions and those obtained by
(5.141).
Figure 5. 17 shows the comparison between the real multi-objective functions results and those
obtained by (5.141).
FIGURE 5. 17: Comparison multi-objective solutions obtained by proposed NSGA-II based

DNN model and their corresponding real values
Then, Figure 5. 18, Figure 5. 19 and Figure 5. 20 illustrate the probability density as a histogram and
∗
the cumulative density of the set of errors 𝒆𝑶𝑳𝑬𝑽
𝟏 for 𝑹𝑬𝑺𝑰𝑶𝑳𝑬𝑽, 𝑻𝑪𝑶𝑳𝑬𝑽 and 𝑨𝑬𝑴𝑰𝑺𝑶𝑳𝑬𝑽,
∗
respectively. Where, 𝒆𝑶𝑳𝑬𝑽
𝟏 denotes the set of relative errors, compared to the max/worst relative
∗
error with 𝑒1𝑂𝐿𝐸𝑉 = max(𝒆𝑶𝑳𝑬𝑽
𝟏 ).
124
∗
𝒆𝑶𝑳𝑬𝑽
𝟏,𝑹𝑬𝑺𝑰 for 𝑹𝑬𝑺𝑰
𝑶𝑳𝑬𝑽
∗
𝟏,𝑻𝑪 for 𝑻𝑪
𝑶𝑳𝑬𝑽
∗
𝟏,𝑨𝑬𝑴𝑰𝑺 for 𝑨𝑬𝑴𝑰𝑺
𝑶𝑳𝑬𝑽
Finally, TABLE 5. 5: Performance of NSGA-IITable 5. 5 summarizes the R2 and errors indices

between both set of solutions. The R2 obtained for the three objective functions are close to 1.00,
𝑒1𝑂𝐿𝐸𝑉 for all objective functions are less than 2%, whereas 𝑒2𝑂𝐿𝐸𝑉 and 𝑒3𝑂𝐿𝐸𝑉 indicate absolute and
relative errors of less than 1%.
TABLE 5. 5: Performance of NSGA-II

Indicators NSGA-II objective functions
𝑂𝐿𝐸𝑉
𝑅𝐸𝑆𝐼 𝑇𝐶 𝑂𝐿𝐸𝑉 𝐴𝐸𝑀𝐼𝑆 𝑂𝐿𝐸𝑉
R2 0.9993 1.000 0.9999
𝑒1𝑂𝐿𝐸𝑉 1.98 (%) 4.71×10-1 (%) 9.45×10-2 (%)
𝑒2𝑂𝐿𝐸𝑉 1.84×10-3 1.57×103 ($/year) 3.94×105 (kg/year)
𝑒3𝑂𝐿𝐸𝑉 5.27×10-1 (%) 1.69×10-2 (%) 1.36×10-2 (%)
125
5.7.3 Optimal Planning Solution
This section presents the optimal planning solution wherein the solution positioned in the pareto
frontier respecting to the NSGA-II behavior. In this case, the solution can be classified into two group
according to the 3D plan for multi-objective functions solutions shown in Figure 5. 21.
GROUP A
GROUP B
FIGURE 5. 21: Classification in Group A and Group B for the multi-objective function’s
solution obtained by NSGA-II based DNN model
Group A corresponds to the solution of 𝑿𝑶𝑳𝑬𝑽 where all three ISWs are installed on candidate
branches of the test system. These solutions present variations due to the allocation of different DDGs
size in the candidate buses and their deployment for maintaining the microgrid operation in islanded
mode. Thus, affecting the evolution of the resilience index in the pareto frontier between ~0.2 and
~0.5. In contrast, Group B correspond to the solution 𝑿𝑶𝑳𝑬𝑽 where none of ISWs are allocated in the
test system. Here, it is worth reminding that a resilience index close to zero represents a high
performing network against severe event or any interruption requirement, i.e., load shedding. In this
perspective, as expected the resilience indices value in Group B are significantly higher in comparison
to Group A.
Detailed results are presented in Table 5. 6 showcasing the planning solution 𝑿𝑶𝑳𝑬𝑽 and multi-
objective functions corresponding to: 1) the max/min value for each objective functions for Group A,
and 2) the two solutions for Group B.
126
TABLE 5. 6: Planning solution corresponding the max/min in each objective function for
group A and the planning solution for group B
Group Binary allocation size of 𝑹𝑬𝑺𝑰𝑶𝑳𝑬𝑽 𝑻𝑪𝑶𝑳𝑬𝑽 𝑨𝑬𝑴𝑰𝑺𝑶𝑳𝑬𝑽
No. DDGs, WTs, PVs and ESSs ($/year) (kg/year)
1 1-4-1 /1-2-1/2-1-2/1-1 0.512 4.53×106 2.85×109
2 10-16-14 /5-5-4/4-4-4/3-3 0.275 1.53×107 2.90×109
7
3 3-30-30 /4-4-4/4-4-4/3-3 0.284 1.78×10 2.89×109
A
4 1-4-1 /1-2-2/1-1-1/1-1 0.506 4.24×106 2.87×109
6
5 8-16-15 /1-2-1/1-1-1/1-1 0.276 9.76×10 3.02×109
6 1-4-2 /3-3-3/4-3-4/2-2 0.478 7.93×106 2.79×109
1 0-2-2 /0-0-0/0-1-1/0-0 0.773 1.53×106 2.86×109
B 6
2 0-2-2 /0-0-0/0-0-0/0-0 0.774 1.03×10 2.88×109
First analyzing Group A, one can observe that cases A2, A3 and A5 show greater resilience in
comparison to A1, A4 and A6. These results are due to the higher size of DDGs allocated, providing
the microgrid with better management capacity for generation/load balancing and, consequently,
avoiding additional curtailment. Nonetheless, increasing the size of DDGs consequently increases the
𝑻𝑪𝑶𝑳𝑬𝑽 and 𝑨𝑬𝑴𝑰𝑺𝑶𝑳𝑬𝑽 due to the additional cost associated. As can see in Table 5. 6, 𝑻𝑪𝑶𝑳𝑬𝑽 and
𝑨𝑬𝑴𝑰𝑺𝑶𝑳𝑬𝑽 for cases A2, A3 and A4 are greater than those in cases A1, A4 and A6. In this sense,
the consideration of renewable resources as main sources leading to a cleaner energy system can
significantly reduce 𝑨𝑬𝑴𝑰𝑺𝑶𝑳𝑬𝑽. This perspective is clearly observed comparing cases A2 to A5,
where 𝑨𝑬𝑴𝑰𝑺𝑶𝑳𝑬𝑽 is significantly reduced for A2 in comparison to A1 due to its higher capacity of
renewable resources. Still, 𝑹𝑬𝑺𝑰𝑶𝑳𝑬𝑽 presents very similar values for both cases with 𝑻𝑪𝑶𝑳𝑬𝑽 for A2
higher in comparison to A5 due to installation cost. On another perspective, Group B solutions feature
significantly reduced resilience capacity, i.e., significantly higher 𝑹𝑬𝑺𝑰𝑶𝑳𝑬𝑽 in comparison to Group
A. However, given their trade-off between the three multi-objective functions, they are still placed
on the pareto frontier, e.g., 𝑻𝑪𝑶𝑳𝑬𝑽 for the cases in Group B are significantly lower in comparison to
A1-A6.
5.7.4 Resilience Improvement
Next, in this section the obtained optimal planning solutions are stressed for a N-3 contingency
scenario to illustrate the system performance in face of severe operating conditions. The N-3
contingency was selected corresponding to the spring season in which multiple faults are localized in
the network, respectively on branches #12, #16 and #19. Following, the network performance
represented by the energy supply level during the N-3 contingency is determined for the base case,
case A1, case A2 and case B2. The corresponding wind gust profile and the binary state representing
127
the availability of branches #12, #16 and #19, according to the selected 𝐀𝐕 𝐈−𝐈𝐈 , are presented in
Figure 5. 22.
FIGURE 5. 22: Wind gust profile and the timeline of branch #12/#16/#19 availability
corresponding to a selected scenario in spring
Whereas Table 5. 7 depicts the availability timeline for branches #12, #16 and #19 along the planning
horizon.
TABLE 5. 7: Timeline of overhead line (branch) #12, #16 and #19 break down over the
planning horizon of a selected scenario in spring
′
line no. 𝒕𝒆 𝒕𝒃𝒓,𝒇𝒂𝒊𝒍𝟎 𝒕𝒅 𝑻𝑻𝑹𝒃𝒓,𝒇𝒂𝒊𝒍 𝒕𝒃𝒓,𝒓𝒆𝒑 𝒕𝒓 𝒕𝒓
8:00 - 11:00 - - 16:00 17:00
#12 9:00 6h 17:00
#16 - 10:00 - 5h 16:00 - -
#19 8:00 5h 16:00
Based on this scenario, energy supply results are illustrated in Figure 5. 23 for the base case and
optimal planning solutions identified in case A1, case A2 and case B2.
Analyzing Figure 5. 23, one can observe that the base case and case B2 are not able to resist to this
severe event, leading to a direct drop of supply from 100% to 0% at 𝑡 𝑒 . As there is no ISWs placed
128
Event start and branch Event stop and Reparation complete and system
#19 fall down start reparation return to grid connected
!
te td tr
100.00
Energy supply level

75.00
50.00
25.00
0.00
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (Hours)
base case A1 A2 B2 Expected level
FIGURE 5. 23: Energy supply level and timeline event for representing system performance
for base case, case A1, case A2 and case B2
23 24 26 27 28 29 30 31 32
23 24 26 27 28 29 30 31 32
25 33
22 25
1 2 12 16
1 2 3 4 5 6 7 8 9 10 11 13 14 15 17
! !
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
18
19
20 21
!
19 20 21 22
(a)
23 24 26 27 28 29 30 31 32
23 24 26 27 28 29 30 31 32
25 33
22
1 2 12 16
1 3 4 5 7 8 9 10 11 13 14 15 17
! !
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
18
19
20 21
!
19 20 21 22
(b)
23 24 26 27 28 29 30 31 32
23 24 26 27 28 29 30 31 32
25 33
22
1 2 12 16
1 3 4 5 7 8 9 10 11 13 14 15 17
! !
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
18
19
20 21
!
19 20 21 22 (c)
23 24 26 27 28 29 30 31 32
23 24 26 27 28 29 30 31 32
25 33
22 25
1 2 12 16
1 2 3 4 5 6 7 8 9 10 11 13 14 15 17
! !
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
18
19
20 21
!
19 20 21 22 (d)
! Fault location due to line broken Bus energized by resources Bus non-energized due to fault(s) or load curtailment
FIGURE 5. 24: Microgrid formations and bus's supply state at 𝐭 𝐝 for (a) Base case (b) Case A1
(c) Case A2 (d) Case B2
129
in the network, the microgrids could not be formed, while only ISW sub found in the substation is
open and take down a whole test system for clearing the faults in this N-3 contingencies. On the other
hand, cases A1 and A2 allow for the formation of microgrids, maintaining the energy supply level
according to the size of DDGs allocated. In this sense, due to the limited size of DDGs allocated in
case A1, the energy supply level drops down to ~20% at 𝑡 𝑑 . In contrast, case A2 shows the best
performance where the energy supply level drops down from 100% to 59% at 𝑡 𝑑 , where 41% of non-
supplied loads correspond to buses located inside locations isolated by ISWs.
5.8 Conclusion
In this work, a novel resilience-oriented microgrid planning framework is proposed. The proposed
framework develops a multi-objective optimization problem formulation incorporating stochastic
DL-LOPFAC models solved by NSGA-II based on DNN. The proposed DL-LOPFAC framework
provides planning solutions considering stochastic scenarios featuring renewable energy and load
uncertainties, HILP events due to N-K contingencies caused by failures in overhead lines, while
simultaneously solving for both GC and IS operation modes, including detailed modeling of control
and operational requirements, such as droop/2nd during islanded. Next, in order to enable this
comprehensive planning framework, NSGA-II based on DNN models is employed to overcome the
computation burden and time execution limitations imposed by the stochastic nature of the resilience-
oriented planning problem and consideration of the detailed systems’ model. In this sense, by taking
advantage of the proposed framework predicted solutions obtained by DL-LOPFAC for the inner level
present a computation efficient solutions that do not require numerous iterations as found in existing
traditional analytical solution methodologies. Following, to verify the effectiveness of the proposed
framework multiple numerical analyses are performed. First, the performance for the DL-LOPFAC is
evaluated by obtaining significantly reduced MSE/RMSE in relation to the expected error threshold.
Moreover, the maximum relative error 𝑒1𝑑𝑛𝑛 is found significantly smaller than the threshold for the
trained DNN models, demonstrating high accuracy for the DNN models’ predicted results. Next, the
proposed optimization level NSGA-II based DNN model is validated, demonstrating accuracy levels
within accepted deviations, specifically the worst relative error 𝑒1𝑂𝐿𝐸𝑉 found is less than 2% for all
fitness functions. Finally, the overall planning solution is discussed, indicating the proposed approach
ability to deliver optimal planning solutions capable of significantly improving the system resilience
even when subjected to HILP events such as N-3 contingencies. Future directions of this work include
the integration of electric vehicles into the optimization problems, as these new system elements can
significantly impact the system operation, while introducing opportunities in the area of resilience-
oriented planning.
130
Conclusion and Perspectives
6.1 Summary and Conclusions

The microgrid concept can offer significant benefits to the power distribution system through both its
grid connected and islanded modes of operation. Given these potential benefits, the microgrid has
been put forward as a key component of the future smart grid. This Ph.D. thesis has presented some
analysis, operation and planning tools that address some of the challenges facing the existing
microgrid power flow tools, optimization models and planning frameworks.
Chapter 2 and Chapter 3 have presented new power flow models needed for the steady state analysis
of practical AC microgrid systems. First, Chapter 2 introduced a non-linear branch-based 𝐙𝐛𝐮𝐬
algorithm for droop controlled islanded microgrids. This algorithm accounts for the distinct features
of the islanded AC microgrid namely; (1) there is no slack bus available for imposing a reference
voltage and frequency, (2) system frequency is not a constant and need to be one of the power flow
variables, and (3) there is as least one DG unit that is being operated as a droop unit. The proposed
algorithm is inversion free and is particularly well suited for sparse implementation. This allows the
proposed algorithm to handle the power flow solution for islanded microgrids created in practical
distribution systems and potentially comprising up to thousands of electrical nodes. The proposed
power flow algorithm has been validated thought several case studies performed on standard balanced
and unbalanced test systems. The numerical results demonstrate the effectiveness of the proposed
algorithm in terms of the accuracy and computational efficiency. Then, Chapter 3 presented a node-
based linear power flow (LPF) model for droop controlled islanded microgrid. Like the algorithm
proposed in Chapter 2, this LPF model accounts for the special features of the AC islanded microgrid
system. The node-based nature of the proposed LPF model allows for its integration in different
planning and operation optimization models to account for the islanded microgrid power flow
operation constraints. The effectiveness of the proposed LPF model has been validated thought
different balanced and unbalanced case studies. The results demonstrated the accuracy of the
proposed LPF model and ensures its suitability for the linear optimal power flow problem
subsequently studied in this thesis.
In chapter 4, the necessary optimal power flow (OPF) models for the optimal operation of AC
microgrids are studied. In this work, linear optimal power flow (LOPF) models for the optimal
operation problem of both balanced and unbalanced AC microgrids with hierarchical control structure
131
were developed. To this end, the optimal operation problem has been first formulated as a MINLP
model, then it was linearized and turned into a MILP model. The proposed MILP model accounts for
the microgrid philosophy of operation during grid connected and islanded through a single
comprehensive model. Several kinds of distributed generations (i.e., dispatchable and non-
dispatchable) and energy storage resources were accounted for in the proposed model. Several
numerical simulation studies were conducted to validate and prove the effectiveness of the proposed
MILP model. The simulation result demonstrated the accuracy of the developed MILP model.
Moreover, the results also reveals that the MILP model’s computation time is significantly lower than
that of its MINLP counterpart model. With such performance enhancement, the proposed LOPF
model was subsequently adopted in the planning problem for accomplishing the main goal of this
thesis.
In the last stage of the research, chapter 5 introduced the microgrid planning framework for resilience
enhancement of the power distribution system when facing the N-K contingencies cause by HILP
event. The proposed framework is based on a multi-objective optimization problem that is solved by
NSGA-II. A deep-learning AC-LOPF (DL-LOPFAC) model was developed to alleviate the
computational burden associated with need to solve the LOPF for multiple candidate solutions and
multiple stochastic scenarios during the evolution of the NSGA-II. In this work, the uncertainty of
the renewable energy has been considered by considering different stochastic scenarios. In order to
account the uncertainty of the N-K contingencies, a methodology for generating severe HILP
operation scenarios depending on the weather has been adopted. The proposed stochastic DL-LOPFAC
accounted for the operation behavior of the microgrids including GC and IS modes of operation,
while properly considering the microgrid’s hierarchical control, by taking advantage of the proposed
LOPF model developed in chapter 4 in the training of the deep learning model. To this end, the
effectiveness of the proposed framework has been proved by numerical simulation results. First, the
performance for the DL-LOPFAC is evaluated and its accuracy is verified. Then, the proposed
optimization level NSGA-II based DNN model is validated and the pareto optimal solution was
derived.
This thesis is mainly conducted through a work presented in chapter 2 to 5 comprising the steady
state analysis (i.e., power flow and optimal power flow carrier) and the planning methodology. The
general contribution of this thesis can be summarized as follows:
• A branch-based inversion free 𝐙𝐛𝐮𝐬 power flow algorithm for droop islanded microgrid is
proposed. The proposed algorithm does not require bus renumbering, nor does it require a
132
DG to be allocated at the reference bus. Moreover, the proposed method is particularly well
suited for sparse implementation and delivers an enhanced computational efficiency in
delivering the power flow solution making it particularly useful for large scale microgrid
systems comprising up to thousands of buses.
• Proposed a node based LPF model that accounts for the droop-controlled autonomous
microgrid philosophy of operation as well as considering both balanced and unbalanced
network characteristics.
• Proposed a time coupled MINLP and MILP OPF models for the optimum operation problem
of balanced and unbalanced microgrid with hierarchical control structure. The proposed
models consider the microgrid philosophy of operation during GC and IS. Moreover, the
energy storage system and the renewable energy resources i.e., WTs and PVs have been
accounted for.
• Proposed a general framework using meta-heuristic multi-objective optimization for planning

resilience oriented microgrid(s) in active distribution systems considering economic,
resilience and environmental points of views. The proposed framework employs novel deep
learning models to speed up the determination of the economic, resilience and environmental
metrics pertaining to the MG(s).
6.2 Perspective for the Future Work

This research has been carried out focusing on the steady-state analysis and the planning
methodology. Future research directions include:
• Explore the new generic power flow and optimal power flow tool that consider multiple and
integrated AC and DC grids.
• Explore other machine learning options and identify different methodologies for comparing
and allowing the planning engineer to choose the appropriate machine learning approach for
the case study under consideration.
• Several types of resources including generators and storages have been considered in the OPF
and planning problems. On the other hand, electric vehicles are increasing in popularity and
will have an impact on the system operation and planning. As such, integrating electric
vehicles is of increasing importance especially in the resilience-oriented planning area.
133
• Explore the transmission level planning, while considering the microgrid at the distribution
level; in the order to the develop the resilience enhancement strategy and planning accounting
for both transmission and distribution systems.
• Test and validate the developed software package to develop an open source software
package suitable for sharing on public repository such as GitHub.
• The studies are conducted using standard test systems. On the other hand, this research can
be adapted and applied by considering the real practical in Laos country. This can lead to the
possibility of exploring new benefit areas, new technologies and the performance
enhancement of the national grid of Laos.
134
Bibliography
[1] Y. Vilaisarn and M. Abdelaziz, “An inversion-free sparse Z power flow algorithm for large-
scale droop controlled islanded microgrids,” Int. J. Electr. Power Energy Syst., vol. 121, no.
August 2019, p. 106048, Oct. 2021, doi: 10.1016/j.ijepes.2020.106048.
[2] Y. Vilaisarn, M. Abdelaziz, and J. Cros, “Linear Three-Phase Power Flow Model for
Autonomous Microgrid Systems,” CIRED - Open Access Proc. J., vol. 2020, no. 1, pp. 50–
53, 2021, doi: 10.1049/oap-cired.2021.0019.
[3] Y. Vilaisarn, M. Moradzadeh, M. Abdelaziz, and J. Cros, “An MILP formulation for the
optimum operation of AC microgrids with hierarchical control,” Int. J. Electr. Power Energy
Syst., vol. 137, no. September 2021, p. 107674, May 2022, doi:
10.1016/j.ijepes.2021.107674.
[4] A. Hirsch, Y. Parag, and J. Guerrero, “Microgrids: A review of technologies, key drivers,
and outstanding issues,” Renewable and Sustainable Energy Reviews, vol. 90, no. April.
Elsevier Ltd, pp. 402–411, 2018, doi: 10.1016/j.rser.2018.03.040.
[5] C. Wang et al., “Distributed Energy and Microgrids (DEM),” Appl. Energy, vol. 210, pp.
685–689, 2018, doi: 10.1016/j.apenergy.2017.11.059.
[6] R. Li, W. Wang, Z. Chen, J. Jiang, and W. Zhang, “A review of optimal planning active
distribution system: Models, methods, and future researches,” Energies, vol. 10, no. 11. pp.
1–27, 2017, doi: 10.3390/en10111715.
[7] S. Parhizi, H. Lotfi, A. Khodaei, and S. Bahramirad, “State of the art in research on
microgrids: A review,” IEEE Access, vol. 3, pp. 890–925, 2015, doi:
10.1109/ACCESS.2015.2443119.
[8] S. Lakshmi and S. Ganguly, “Transition of Power Distribution System Planning from
Passive to Active Networks: A State-of-the-Art Review and a New Proposal,” in Green
Energy and Technology, no. 9789811071874, 2018, pp. 87–117.
[9] C. A. C. D’ADAMO, S. JUPE, “Global Survey on Planning and Operation of Active

Distribution Network-Update of CIGRE C6.11 Working Group Activities,” 2009.
[10] R. Hidalgo, C. Abbey, and G. Joós, “A review of active distribution networks enabling
135
technologies,” in IEEE PES General Meeting, PES 2010, 2010, pp. 1–9, doi:
10.1109/PES.2010.5590051.
[11] D. T. Ton and M. A. Smith, “The U.S. Department of Energy’s Microgrid Initiative,” Electr.
J., vol. 25, no. 8, pp. 84–94, 2012, doi: 10.1016/j.tej.2012.09.013.
[12] R. H. Lasseter, “MicroGrids,” in 2002 IEEE Power Engineering Society Winter Meeting.
Conference Proceedings (Cat. No.02CH37309), 2002, vol. 1, pp. 305–308, doi:
10.1109/PESW.2002.985003.
[13] E. Bullich-Massagué, F. Díaz-González, M. Aragüés-Peñalba, F. Girbau-Llistuella, P.

Olivella-Rosell, and A. Sumper, “Microgrid clustering architectures,” Appl. Energy, vol.
212, no. August 2017, pp. 340–361, 2018, doi: 10.1016/j.apenergy.2017.12.048.
[14] D. E. Olivares et al., “Trends in microgrid control,” IEEE Trans. Smart Grid, vol. 5, no. 4,
pp. 1905–1919, Jul. 2014, doi: 10.1109/TSG.2013.2295514.
[15] M. M. A. Abdelaziz, H. E. Farag, E. F. El-Saadany, and Y. A.-R. I. Mohamed, “A Novel

and Generalized Three-Phase Power Flow Algorithm for Islanded Microgrids Using a
Newton Trust Region Method,” IEEE Trans. Power Syst., vol. 28, no. 1, pp. 190–201, Feb.
2013, doi: 10.1109/TPWRS.2012.2195785.
[16] G. Agundis-Tinajero, J. Segundo-Ramírez, N. Visairo-Cruz, M. Savaghebi, J. M. Guerrero,

and E. Barocio, “Power flow modeling of islanded AC microgrids with hierarchical control,”
Int. J. Electr. Power Energy Syst., vol. 105, no. August 2018, pp. 28–36, Feb. 2019, doi:
10.1016/j.ijepes.2018.08.002.
[17] M. H. Moradi, V. B. Foroutan, and M. Abedini, “Power flow analysis in islanded Micro-
Grids via modeling different operational modes of DGs: A review and a new approach,”
Renew. Sustain. Energy Rev., vol. 69, no. August 2015, pp. 248–262, Mar. 2017, doi:
10.1016/j.rser.2016.11.156.
[18] J. D. Glover, M. S. Sarma, and T. J. Overbye, Power System Analysis and Design. 2012.
[19] Turan Gonen, Modern Power System Analysis. 2016.
[20] O. A. Afolabi, W. H. Ali, P. Cofie, J. Fuller, P. Obiomon, and E. S. Kolawole, “Analysis of

the Load Flow Problem in Power System Planning Studies,” Energy Power Eng., vol. 07,
136
no. 10, pp. 509–523, 2015, doi: 10.4236/epe.2015.710048.
[21] S. Sunisith and K. Meena, “Backward / Forward Sweep Based Distribution Load Flow
Method,” in International Electrical Engineering Journal, 2014, vol. 5, no. 9, pp. 1539–
1544.
[22] P. J. Lagace, M. H. Vuong, and I. Kamwa, “Improving power flow convergence by Newton
Raphson with a levenberg-marquardt method,” IEEE Power Energy Soc. 2008 Gen. Meet.
Convers. Deliv. Electr. Energy 21st Century, PES, pp. 1–6, 2008, doi:
10.1109/PES.2008.4596138.
[23] G. Agundis-Tinajero et al., “Extended-Optimal-Power-Flow-Based Hierarchical Control for

Islanded AC Microgrids,” IEEE Trans. Power Electron., vol. 34, no. 1, pp. 840–848, Jan.
2019, doi: 10.1109/TPEL.2018.2813980.
[24] H. Abdi, S. D. Beigvand, and M. La Scala, “A review of optimal power flow studies applied
to smart grids and microgrids,” Renew. Sustain. Energy Rev., vol. 71, no. December 2016,
pp. 742–766, 2017, doi: 10.1016/j.rser.2016.12.102.
[25] A. Vaccaro and C. A. Cañizares, “A knowledge-based framework for power flow and
optimal power flow analyses,” IEEE Trans. Smart Grid, vol. 9, no. 1, pp. 230–239, 2018,
doi: 10.1109/TSG.2016.2549560.
[26] P. P. Vergara, J. C. Lopez, M. J. Rider, and L. C. P. Da Silva, “Optimal Operation of

Unbalanced Three-Phase Islanded Droop-Based Microgrids,” IEEE Trans. Smart Grid, vol.
10, no. 1, pp. 928–940, 2019, doi: 10.1109/TSG.2017.2756021.
[27] S. A. Arefifar, Y. A.-R. I. Mohamed, and T. H. M. El-Fouly, “Supply-Adequacy-Based

Optimal Construction of Microgrids in Smart Distribution Systems,” IEEE Trans. Smart
Grid, vol. 3, no. 3, pp. 1491–1502, Sep. 2012, doi: 10.1109/TSG.2012.2198246.
[28] S. A. Arefifar, Y. A.-R. I. Mohamed, and T. H. M. EL-Fouly, “Optimum Microgrid Design

for Enhancing Reliability and Supply-Security,” IEEE Trans. Smart Grid, vol. 4, no. 3, pp.
1567–1575, Sep. 2013, doi: 10.1109/TSG.2013.2259854.
[29] M. Ghasemi, A. Kazemi, A. Mazza, and E. Bompard, “A three-stage stochastic planning

model for enhancing the resilience of distribution systems with microgrid formation
137
strategy,” IET Gener. Transm. Distrib., vol. 15, no. 13, pp. 1908–1921, 2021, doi:
10.1049/gtd2.12144.
[30] F. Hameed, M. Al Hosani, and H. H. Zeineldin, “A Modified Backward/Forward Sweep

Load Flow Method for Islanded Radial Microgrids,” IEEE Trans. Smart Grid, vol. 10, no. 1,
pp. 910–918, Jan. 2019, doi: 10.1109/TSG.2017.2754551.
[31] F. Mumtaz, M. H. Syed, M. Al Hosani, and H. H. Zeineldin, “A Novel Approach to Solve

Power Flow for Islanded Microgrids Using Modified Newton Raphson with Droop Control
of DG,” IEEE Trans. Sustain. Energy, vol. 7, no. 2, pp. 493–503, 2016, doi:
10.1109/TSTE.2015.2502482.
[32] A. Kumar, B. K. Jha, D. Singh, and R. K. Misra, “Current injection-based Newton–Raphson

power-flow algorithm for droop-based islanded microgrids,” IET Gener. Transm. Distrib.,
vol. 13, no. 23, pp. 5271–5283, Dec. 2019, doi: 10.1049/iet-gtd.2019.0575.
[33] A. A. Nazari, R. Keypour, M. H. Beiranvand, and N. Amjady, “A decoupled extended

power flow analysis based on Newton-Raphson method for islanded microgrids,” Int. J.
Electr. Power Energy Syst., vol. 117, no. September 2019, p. 105705, 2020, doi:
10.1016/j.ijepes.2019.105705.
[34] A. Kumar, B. K. Jha, D. K. Dheer, R. K. Misra, and D. Singh, “A Nested-Iterative Newton-

Raphson based Power Flow Formulation for Droop-based Islanded Microgrids,” Electr.
Power Syst. Res., vol. 180, no. December 2019, p. 106131, 2020, doi:
10.1016/j.epsr.2019.106131.
[35] G. Diaz, J. Gomez-Aleixandre, and J. Coto, “Direct Backward/Forward Sweep Algorithm

for Solving Load Power Flows in AC Droop-Regulated Microgrids,” IEEE Trans. Smart
Grid, vol. 7, no. 5, pp. 2208–2217, Sep. 2016, doi: 10.1109/TSG.2015.2478278.
[36] M. E. Nassar and M. M. A. Salama, “A novel branch-based power flow algorithm for
islanded AC microgrids,” Electr. Power Syst. Res., vol. 146, pp. 51–62, May 2017, doi:
10.1016/j.epsr.2017.01.019.
[37] A. Kumar, B. K. Jha, D. K. Dheer, D. Singh, and R. K. Misra, “Nested backward/forward

sweep algorithm for power flow analysis of droop regulated islanded microgrids,” IET
Gener. Transm. Distrib., vol. 13, no. 14, pp. 3086–3095, 2019, doi: 10.1049/iet-
138
gtd.2019.0388.
[38] G. C. Kryonidis, E. O. Kontis, A. I. Chrysochos, K. O. Oureilidis, C. S. Demoulias, and G.

K. Papagiannis, “Power Flow of Islanded AC Microgrids: Revisited,” IEEE Trans. Smart
Grid, vol. 9, no. 4, pp. 3903–3905, 2018, doi: 10.1109/TSG.2018.2799480.
[39] T. Y. Hsieh, T. H. Chen, and N. C. Yang, “Matrix decompositions-based approach to Z-bus

matrix building process for radial distribution systems,” Int. J. Electr. Power Energy Syst.,
vol. 89, pp. 62–68, 2017, doi: 10.1016/j.ijepes.2017.01.004.
[40] N. C. Yang, “Three-phase power flow calculations using direct ZBUS method for large-
scale unbalanced distribution networks,” IET Gener. Transm. Distrib., vol. 10, no. 4, pp.
1048–1055, 2016, doi: 10.1049/iet-gtd.2015.0822.
[41] J. H. Teng, “A direct approach for distribution system load flow solutions,” IEEE Trans.
Power Deliv., vol. 18, no. 3, pp. 882–887, 2003, doi: 10.1109/TPWRD.2003.813818.
[42] P. Ling, C. Fang, L. Guo, X. Su, and S. Zheng, “A Novel Direct Load Flow Algorithm for
Unbalanced Micro-grids Considering the Droop Characteristics of DG and Load,” 2018 Int.
Conf. Power Syst. Technol. POWERCON 2018 - Proc., no. 201805280000069, pp. 2096–
2101, 2019, doi: 10.1109/POWERCON.2018.8601869.
[43] X. Wang, M. Shahidehpour, C. Jiang, W. Tian, Z. Li, and Y. Yao, “Three-Phase Distribution
Power Flow Calculation for Loop-Based Microgrids,” IEEE Trans. Power Syst., vol. 33, no.
4, pp. 3955–3967, 2018, doi: 10.1109/TPWRS.2017.2788055.
[44] E. E. Pompodakis, G. C. Kryonidis, and M. C. Alexiadis, “A Comprehensive Load Flow

Approach for Grid-Connected and Islanded AC Microgrids,” IEEE Trans. Power Syst., vol.
35, no. 2, pp. 1143–1155, Mar. 2020, doi: 10.1109/TPWRS.2019.2945011.
[45] M. Abdelaziz, “GPU-OpenCL accelerated probabilistic power flow analysis using Monte-
Carlo simulation,” Electr. Power Syst. Res., vol. 147, pp. 70–72, Jun. 2017, doi:
10.1016/j.epsr.2017.02.022.
[46] Y. Wang, N. Zhang, H. Li, J. Yang, and C. Kang, “Linear three-phase power flow for
unbalanced active distribution networks with PV nodes,” CSEE J. Power Energy Syst., vol.
3, no. 3, pp. 321–324, 2017, doi: 10.17775/cseejpes.2017.00240.
139
[47] Z. Yang, K. Xie, J. Yu, H. Zhong, N. Zhang, and Q. X. Xia, “A General Formulation of
Linear Power Flow Models: Basic Theory and Error Analysis,” IEEE Trans. Power Syst.,
vol. 34, no. 2, pp. 1315–1324, 2019, doi: 10.1109/TPWRS.2018.2871182.
[48] M. M. A. Abdelaziz and H. E. Farag, “An Enhanced Supervisory Control for Islanded
Microgrid Systems,” IEEE Trans. Smart Grid, vol. 7, no. 4, pp. 1941–1943, 2016, doi:
10.1109/TSG.2016.2553580.
[49] J. Rocabert, A. Luna, F. Blaabjerg, and P. Rodríguez, “Control of power converters in AC

microgrids,” IEEE Trans. Power Electron., vol. 27, no. 11, pp. 4734–4749, 2012, doi:
10.1109/TPEL.2012.2199334.
[50] D. Wu, F. Tang, T. Dragicevic, J. C. Vasquez, and J. M. Guerrero, “A Control Architecture

to Coordinate Renewable Energy Sources and Energy Storage Systems in Islanded
Microgrids,” IEEE Trans. Smart Grid, vol. 6, no. 3, pp. 1156–1166, 2015, doi:
10.1109/TSG.2014.2377018.
[51] X. Wu, X. Wang, and C. Qu, “A hierarchical framework for generation scheduling of
microgrids,” IEEE Trans. Power Deliv., vol. 29, no. 6, pp. 2448–2457, 2014, doi:
10.1109/TPWRD.2014.2360064.
[52] P. P. Vergara, J. C. López, L. C. P. da Silva, and M. J. Rider, “Security-constrained optimal

energy management system for three-phase residential microgrids,” Electr. Power Syst. Res.,
vol. 146, pp. 371–382, 2017, doi: 10.1016/j.epsr.2017.02.012.
[53] E. Riva Sanseverino, N. Nguyen Quang, M. L. Di Silvestre, J. M. Guerrero, and C. Li,

“Optimal power flow in three-phase islanded microgrids with inverter interfaced units,”
Electr. Power Syst. Res., vol. 123, pp. 48–56, 2015, doi: 10.1016/j.epsr.2015.01.020.
[54] P. P. Vergara et al., “A Generalized Model for the Optimal Operation of Microgrids in Grid-
Connected and Islanded Droop-Based Mode,” IEEE Trans. Smart Grid, vol. 10, no. 5, pp.
5032–5045, 2018, doi: 10.1109/TSG.2018.2873411.
[55] F. H. Jufri, V. Widiputra, and J. Jung, “State-of-the-art review on power grid resilience to
extreme weather events: Definitions, frameworks, quantitative assessment methodologies,
and enhancement strategies,” Appl. Energy, vol. 239, no. February, pp. 1049–1065, 2019,
doi: 10.1016/j.apenergy.2019.02.017.
140
[56] Z. Li, M. Shahidehpour, F. Aminifar, A. Alabdulwahab, and Y. Al-Turki, “Networked
Microgrids for Enhancing the Power System Resilience,” Proc. IEEE, vol. 105, no. 7, pp.
1289–1310, 2017, doi: 10.1109/JPROC.2017.2685558.
[57] Y. Wang, A. O. Rousis, and G. Strbac, “On microgrids and resilience: A comprehensive
review on modeling and operational strategies,” Renew. Sustain. Energy Rev., vol. 134, no.
November 2019, p. 110313, 2020, doi: 10.1016/j.rser.2020.110313.
[58] J. Kim and Y. Dvorkin, “Enhancing Distribution System Resilience with Mobile Energy
Storage and Microgrids,” IEEE Trans. Smart Grid, vol. 10, no. 5, pp. 4996–5006, 2018, doi:
10.1109/TSG.2018.2872521.
[59] Y. Wang, A. Oulis-Rousis, and G. Strbac, “A three-level planning model for optimal sizing
of networked microgrids considering a trade-off between resilience and cost,” IEEE Trans.
Power Syst., vol. 36, no. 6, pp. 5657–5669, 2021, doi: 10.1109/TPWRS.2021.3076128.
[60] J. Najafi, A. Peiravi, A. Anvari-Moghaddam, and J. M. Guerrero, “Resilience improvement

planning of power-water distribution systems with multiple microgrids against hurricanes
using clean strategies,” J. Clean. Prod., vol. 223, pp. 109–126, 2019, doi:
10.1016/j.jclepro.2019.03.141.
[61] M. H. Amirioun, F. Aminifar, and H. Lesani, “Towards Proactive Scheduling of Microgrids

Against Extreme Floods,” IEEE Trans. Smart Grid, vol. 9, no. 4, pp. 3900–3902, 2018, doi:
10.1109/TSG.2017.2762906.
[62] Q. Shi et al., “Network reconfiguration and distributed energy resource scheduling for
improved distribution system resilience,” Int. J. Electr. Power Energy Syst., vol. 124, no.
July 2020, p. 106355, 2021, doi: 10.1016/j.ijepes.2020.106355.
[63] J. Zhu, Y. Yuan, and W. Wang, “An exact microgrid formation model for load restoration in
resilient distribution system,” Int. J. Electr. Power Energy Syst., vol. 116, no. September
2019, p. 105568, 2020, doi: 10.1016/j.ijepes.2019.105568.
[64] X. Wu, Z. Wang, T. Ding, X. Wang, Z. Li, and F. Li, “Microgrid planning considering the
resilience against contingencies,” IET Gener. Transm. Distrib., vol. 13, no. 16, pp. 3534–
3548, 2019, doi: 10.1049/iet-gtd.2018.6816.
141
[65] Y. Wang, A. O. Rousis, and G. Strbac, “Resilience-driven optimal sizing and pre-positioning
of mobile energy storage systems in decentralized networked microgrids,” Appl. Energy,
vol. 305, no. September 2021, p. 117921, 2022, doi: 10.1016/j.apenergy.2021.117921.
[66] X. Hu, H. Hu, S. Verma, and Z.-L. Zhang, “Physics-Guided Deep Neural Networks for
Power Flow Analysis,” IEEE Trans. Power Syst., vol. 36, no. 3, pp. 1–1, 2020, doi:
10.1109/TPWRS.2020.3029557.
[67] X. Pan, T. Zhao, M. Chen, and S. Zhang, “DeepOPF: A Deep Neural Network Approach for
Security-Constrained DC Optimal Power Flow,” IEEE Trans. Power Syst., vol. 36, no. 3, pp.
1–1, 2020, doi: 10.1109/TPWRS.2020.3026379.
[68] J. C. Bedoya, Y. Wang, and C. Liu, “Distribution System Resilience Under Asynchronous
Information Using Deep Reinforcement Learning Juan,” IEEE Trans. Power Syst., vol. 36,
no. 5, pp. 4235–4245, 2021.
[69] M. Moradi-Sepahvand, T. Amraee, and S. Sadeghi Gougheri, “Deep Learning Based

Hurricane Resilient Coplanning of Transmission Lines, Battery Energy Storages, and Wind
Farms,” IEEE Trans. Ind. Informatics, vol. 18, no. 3, pp. 2120–2131, 2022, doi:
10.1109/TII.2021.3074397.
[70] M. Dabbaghjamanesh, A. Kavousi-Fard, and J. Zhang, “Stochastic Modeling and Integration

of Plug-In Hybrid Electric Vehicles in Reconfigurable Microgrids with Deep Learning-
Based Forecasting,” IEEE Trans. Intell. Transp. Syst., vol. 22, no. 7, pp. 4394–4403, 2021,
doi: 10.1109/TITS.2020.2973532.
[71] M. M. A. Abdelaziz, “OpenCL-Accelerated Probabilistic Power Flow for Active

Distribution Networks,” IEEE Trans. Sustain. Energy, vol. 9, no. 3, pp. 1255–1264, Jul.
2018, doi: 10.1109/TSTE.2017.2781148.
[72] M. A. Allam, A. A. Hamad, M. Kazerani, and E. F. El-Saadany, “A novel dynamic power

routing scheme to maximize loadability of islanded hybrid AC/DC microgrids under
unbalanced AC loading,” IEEE Trans. Smart Grid, vol. 9, no. 6, pp. 5798–5809, 2018, doi:
10.1109/TSG.2017.2697360.
[73] “C. Ababei, REDS: REpository of Distribution Systems, 2012. [Online]. Available:
http://www.dejazzer.com Accessed on: Aug. 11, 2016.” .
142
[74] G. K. V. Raju and P. R. Bijwe, “Efficient reconfiguration of balanced and unbalanced
distribution systems for loss minimisation,” Gener. Transm. Distrib. IET, vol. 2, no. 1, pp.
7–12, doi: 10.1049/iet-gtd.
[75] M. M. A. Abdelaziz, H. E. Z. Farag, and E. F. El-Saadany, “Fuel-saving benefit analysis of

islanded microgrid central controllers,” Electr. Power Syst. Res., vol. 125, pp. 45–54, 2015,
doi: 10.1016/j.epsr.2015.03.026.
[76] M. M. A. Abdelaziz, “Effect of Detailed Reactive Power Limit Modeling on Islanded

Microgrid Power Flow Analysis,” IEEE Trans. Power Syst., vol. 31, no. 2, pp. 1665–1666,
2016, doi: 10.1109/TPWRS.2015.2412690.
[77] H. E. Farag, M. M. A. Abdelaziz, and E. F. El-Saadany, “Incorporating voltage regulator

and load models in unbalanced power flow studies of active distribution systems,” IEEE
Power Energy Soc. Gen. Meet., 2012, doi: 10.1109/PESGM.2012.6344972.
[78] M. Abdelaziz and M. Moradzadeh, “Monte-Carlo simulation based multi-objective optimum

allocation of renewable distributed generation using OpenCL,” Electr. Power Syst. Res., vol.
170, no. September 2018, pp. 81–91, May 2019, doi: 10.1016/j.epsr.2019.01.012.
[79] M. M. A. Abdelaziz, H. E. Farag, and E. F. El-Saadany, “Optimum Reconfiguration of

Droop-Controlled Islanded Microgrids,” IEEE Trans. Power Syst., vol. 31, no. 3, pp. 2144–
2153, May 2016, doi: 10.1109/TPWRS.2015.2456154.
[80] J. M. Guerrero, M. Chandorkar, T. L. Lee, and P. C. Loh, “Advanced control architectures

for intelligent microgridspart i: Decentralized and hierarchical control,” IEEE Trans. Ind.
Electron., vol. 60, no. 4, pp. 1254–1262, 2013, doi: 10.1109/TIE.2012.2194969.
[81] L. H. Macedo, S. Member, J. F. Franco, and M. J. Rider, “Optimal Operation of Distribution

Networks Considering Energy Storage Devices,” IEEE Trans. Smart Grid, vol. 6, no. 6, pp.
1–12, 2015.
[82] S. M. Mohseni-Bonab, I. Kamwa, A. Moeini, and A. Rabiee, “Voltage Security Constrained

Stochastic Programming Model for Day-Ahead BESS Schedule in Co-Optimization of T&D
Systems,” IEEE Trans. Sustain. Energy, vol. 11, no. 1, pp. 391–404, 2020, doi:
10.1109/TSTE.2019.2892024.
143
[83] M. Moradzadeh and M. M. A. Abdelaziz, “A New MILP Formulation for Renewables and
Energy Storage Integration in Fast Charging Stations,” IEEE Trans. Transp. Electrif., vol. 6,
no. 1, pp. 181–198, 2020, doi: 10.1109/TTE.2020.2974179.
[84] I. Alsaidan, A. Khodaei, and W. Gao, “A Comprehensive Battery Energy Storage Optimal
Sizing Model for Microgrid Applications,” IEEE Trans. Power Syst., vol. 33, no. 4, pp.
3968–3980, Jul. 2018, doi: 10.1109/TPWRS.2017.2769639.
[85] M. Moradzadeh and M. Abdelaziz, “A Stochastic Optimal Planning Model for Fully Green
Stand-Alone PEV Charging Stations,” IEEE Trans. Transp. Electrif., 2021, doi:
10.1109/TTE.2021.3069438.
[86] Z. Yang, H. Zhong, Q. Xia, A. Bose, and C. Kang, “Optimal power flow based on
successive linear approximation of power flow equations,” IET Gener. Transm. Distrib., vol.
10, no. 14, pp. 3654–3662, Nov. 2016, doi: 10.1049/iet-gtd.2016.0547.
[87] Z. Yang, A. Bose, H. Zhong, N. Zhang, Q. Xia, and C. Kang, “Optimal Reactive Power
Dispatch With Accurately Modeled Discrete Control Devices: A Successive Linear
Approximation Approach,” IEEE Trans. Power Syst., vol. 32, no. 3, pp. 2435–2444, May
2017, doi: 10.1109/TPWRS.2016.2608178.
[88] Z. Yang, H. Zhong, A. Bose, T. Zheng, Q. Xia, and C. Kang, “A Linearized OPF Model
with Reactive Power and Voltage Magnitude: A Pathway to Improve the MW-Only DC
OPF,” IEEE Trans. Power Syst., vol. 33, no. 2, pp. 1734–1745, 2018, doi:
10.1109/TPWRS.2017.2718551.
[89] “New York Independent System Operator. [Online]. Available: http://www.nyiso.com.” .
[90] M. Barani, J. Aghaei, M. A. Akbari, T. Niknam, H. Farahmand, and M. Korpas, “Optimal

Partitioning of Smart Distribution Systems Into Supply-Sufficient Microgrids,” IEEE Trans.
Smart Grid, vol. 10, no. 3, pp. 2523–2533, May 2019, doi: 10.1109/TSG.2018.2803215.
[91] F. S. Gazijahani and J. Salehi, “Robust Design of Microgrids with Reconfigurable Topology
under Severe Uncertainty,” IEEE Trans. Sustain. Energy, vol. 9, no. 2, pp. 559–569, 2018,
doi: 10.1109/TSTE.2017.2748882.
[92] F. Samadi Gazijahani and J. Salehi, “Optimal Bilevel Model for Stochastic Risk-Based
144
Planning of Microgrids Under Uncertainty,” IEEE Trans. Ind. Informatics, vol. 14, no. 7, pp.
3054–3064, Jul. 2018, doi: 10.1109/TII.2017.2769656.
[93] A. J. Conejo, M. Carrion, and J. M. Morales, Decision making under uncertainty in

electricity markets. IEEE, 2010.
[94] M. Bao, Y. Ding, M. Sang, D. Li, C. Shao, and J. Yan, “Modeling and evaluating nodal
resilience of multi-energy systems under windstorms,” Appl. Energy, vol. 270, no. January,
p. 115136, 2020, doi: 10.1016/j.apenergy.2020.115136.
[95] M. H. Amirioun, F. Aminifar, and H. Lesani, “Resilience-Oriented Proactive Management

of Microgrids Against Windstorms,” IEEE Trans. Power Syst., vol. 33, no. 4, pp. 4275–
4284, 2018, doi: 10.1109/TPWRS.2017.2765600.
[96] M. H. Amirioun, F. Aminifar, H. Lesani, and M. Shahidehpour, “Metrics and quantitative

framework for assessing microgrid resilience against windstorms,” Int. J. Electr. Power
Energy Syst., vol. 104, no. January 2018, pp. 716–723, 2019, doi:
10.1016/j.ijepes.2018.07.025.
[97] J. W. Muhs and M. Parvania, “Stochastic Spatio-Temporal Hurricane Impact Analysis for
Power Grid Resilience Studies,” 2019, doi: 10.1109/ISGT.2019.8791647.
[98] Quanta Technology, “Cost-Benefit Analysis of the Deployment of Utility Infrastructure

Upgrades and Storm Hardening Programs,” Satell. Commun., vol. 3021, pp. 1–108, 2009.
[99] Y. M. Darestani and A. Shafieezadeh, “Multi-dimensional wind fragility functions for wood
utility poles,” Eng. Struct., vol. 183, no. August 2018, pp. 937–948, 2019, doi:
10.1016/j.engstruct.2019.01.048.
[100] T. A. Short, Electric Power Distribution Equipment and Systems. New York, 2006.
[101] S. Poudel, A. Dubey, and A. Bose, “Risk-based probabilistic quantification of power

distribution system operational resilience,” IEEE Syst. J., vol. 14, no. 3, pp. 3506–3517,
2019.
[102] B. Mukhopadhyay and D. Das, “Optimal multi-objective expansion planning of a droop-

regulated islanded microgrid,” Energy, vol. 218, p. 119415, 2021, doi:
10.1016/j.energy.2020.119415.
145
[103] M. F. Shaaban and E. F. El-Saadany, “Accommodating high penetrations of pevs and
renewable dg considering uncertainties in distribution systems,” IEEE Trans. Power Syst.,
vol. 29, no. 1, pp. 259–270, 2014, doi: 10.1109/TPWRS.2013.2278847.
[104] A. Maulik and D. Das, “Optimal Operation of Droop-Controlled Islanded Microgrids,” IEEE
Trans. Sustain. Energy, vol. 9, no. 3, pp. 1337–1348, Jul. 2018, doi:
10.1109/TSTE.2017.2783356.
[105] J. David, M. Van Nguyen, and N. Cohen, “Algorithmic Graph Theory,” 2012.
[106] R. Balakrishnan and K. Ranganathan, A Textbook of Graph Theory. New York, NY:
Springer New York, 2012.
[107] H. E. Farag, M. M. A. Abdelaziz, and E. F. El-Saadany, “Voltage and reactive power

impacts on successful operation of islanded microgrids,” IEEE Trans. Power Syst., vol. 28,
no. 2, pp. 1716–1727, 2013, doi: 10.1109/TPWRS.2012.2223491.
[108] M. M. A. Abdelaziz, H. E. Farag, and E. F. El-Saadany, “Optimum droop parameter settings

of islanded microgrids with renewable energy resources,” IEEE Trans. Sustain. Energy, vol.
5, no. 2, pp. 434–445, 2014, doi: 10.1109/TSTE.2013.2293201.
[109] R. Tarjan, “Depth-First Search and Linear Graph Algorithms,” SIAM J. Comput., vol. 1, no.
2, pp. 146–160, Jun. 1972, doi: 10.1137/0201010.
[110] C. S. Thomas H. Cormen, Charles E. Leiserson, Ronald L. Rivest, Introduction to

Algorithms, Third Edition. 2009.
[111] H. and B. Hagan M. T., Demuth, Neural Network Design (2nd Edition). Boston, 2014.
[112] S. Haykin, Neural Networks and Learning, vol. 1–3. Elsevier, 2008.
[113] M. H. Beale, M. T. Hagan, and H. B. Demuth, Deep Learning ToolboxTM User’s Gude.
2021.
[114] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist multiobjective genetic
algorithm: NSGA-II,” IEEE Trans. Evol. Comput., vol. 6, no. 2, pp. 182–197, 2002, doi:
10.1109/4235.996017.
[115] “The National Renewable Energy Laboratory (NREL). [Onlne]. Available:
146
https://maps.nrel.gov/nsrdb-viewer.” .
[116] H. Haddadian and R. Noroozian, “Multi-microgrids approach for design and operation of
future distribution networks based on novel technical indices,” Appl. Energy, vol. 185, pp.
650–663, 2017, doi: 10.1016/j.apenergy.2016.10.120.
[117] L. Wang, A. H. C. Ng, and K. Deb, Multi-objective Evolutionary Optimisation for Product
Design and Manufacturing. London: Springer London, 2011.
[118] Dimitris Bertsimas, Introduction to Linear Optimization. 1997.
[119] “U.S. Energy Information Administration. [Online]. Available:

https://www.eia.gov/dnav/ng/hist/n3035us3m.htm,” 2020. .
147
Appendix A: Element of square matrix
The elements of the square matrix (2.30) were derived by the partial derivative of the power flow
equations (2.44) and (2.45) over 𝑉 and 𝛿.
𝛹(𝛽) 𝛹𝛽
𝑛
|𝑉̅𝑖,𝛽 | |𝑌𝑖𝑗 | cos (𝜃𝑖𝑗 +𝛿𝑖,𝛽 − 𝛿𝑖,𝛹 )
𝑃𝑖,𝛹 = |𝑉̅𝑖,𝛹 | ∑𝑗=1 ∑𝛽∈ℱ [
𝑏𝑢𝑠
𝛹(𝛽) 𝛹𝛽
]
𝑗≠𝑖 −|𝑉𝑗,𝛽 | |𝑌𝑖𝑗 | cos (𝜃𝑖𝑗 +𝛿𝑗,𝛽 − 𝛿𝑖,𝛹 )
,𝛹 ∈ ℱ; ∀𝑖 ∈ 𝕯 (2.44)
𝛹(𝛽) 𝛹𝛽
𝑛𝑏𝑢𝑠
|𝑉𝑗,𝛽 | |𝑌𝑖𝑗 | sin (𝜃𝑖𝑗 +𝛿𝑗,𝛽 − 𝛿𝑖,𝛹 )
𝑃𝑖,𝛹 = |𝑉̅𝑖,𝛹 | ∑𝑗=1 ∑𝛽∈ℱ [ 𝛹(𝛽) 𝛹𝛽
]
𝑗≠𝑖 −|𝑉̅𝑖,𝛽 | |𝑌𝑖𝑗 | sin (𝜃𝑖𝑗 +𝛿𝑖,𝛽 − 𝛿𝑖,𝛹 )
,𝛹 ∈ ℱ; ∀𝑖 ∈ 𝕯 (2.44)
In this work, the branch admittance was computed elements by elements thought the corresponding
elements of 𝐙𝐛𝐫,𝟑Ф expressed by (2.46). Hence, the formulation 𝐘 𝐛𝐮𝐬 is unneeded. Then, (2.47) is the
expression of branch admittance using in this work.
𝑓3Ф
𝐙𝐛𝐫,𝟑Ф = 𝑟𝑒𝑎𝑙(𝐙𝐛𝐫,𝟑Ф ) + 𝑖𝑚𝑎𝑔(𝐙𝐛𝐫,𝟑Ф ) 𝑓0
(2.46)
𝛹𝛽 𝛹𝛽 1
|𝑌𝑖𝑗 | ∠𝜃𝑖𝑗 = 𝑏𝑟,𝛹𝛽 (2.47)
𝑍̅𝑖𝑗
According to the active and reactive power injection delivered from the inner loop, the equation (2.48)
to (2.55) illustrated the simplification partial derivative models of (2.44) and (2.45) depend on 𝑃𝑖𝑎,𝑏,𝑐
and 𝑄𝑖𝑎,𝑏,𝑐 :
𝜕𝑃𝑖𝑏 𝑛
= −𝑄𝑖𝑏 + ∑𝑗=2
𝑏𝑢𝑠
|𝑌𝑖𝑗𝑏𝑏 ||𝑉𝑖𝑏 |(cos𝜃𝑖𝑗
𝑏𝑏 𝑏𝑏
− 𝑠𝑖𝑛𝜃𝑖𝑗 ) (2.48)
𝜕𝛿𝑖𝑏
𝜕𝑃𝑖𝑐 𝑛
𝜕𝛿𝑖𝑐
= −𝑄𝑖𝑐 + ∑𝑗=2
𝑏𝑢𝑠
|𝑌𝑖𝑗𝑐𝑐 ||𝑉𝑖𝑐 |(cos𝜃𝑖𝑗
𝑐𝑐 𝑐𝑐
− 𝑠𝑖𝑛𝜃𝑖𝑗 ) (2.49)
𝜕𝑃𝑖𝑏 𝑃𝑖𝑏 𝑛
= 𝑏𝑢𝑠
+ ∑𝑗=2 |𝑌𝑖𝑗𝑏𝑏 ||𝑉𝑖𝑏 |𝑐𝑜𝑠𝜃𝑖𝑗
𝑏𝑏
(2.50)
𝜕𝑉𝑖𝑏 |𝑉𝑖𝑏 |
𝜕𝑃𝑖𝑐 𝑃𝑖𝑐 𝑛
𝜕𝑉𝑖𝑐
=
|𝑉𝑖𝑐 |
𝑏𝑢𝑠
+ ∑𝑗=2 |𝑌𝑖𝑗𝑐𝑐 ||𝑉𝑖𝑐 |𝑐𝑜𝑠𝜃𝑖𝑗
𝑐𝑐
(2.51)
148
𝜕𝑄𝑖𝑏 𝑛
= 𝑃𝑖𝑏 + ∑𝑗=2
𝑏𝑢𝑠
−|𝑌𝑖𝑗𝑏𝑏 ||𝑉𝑖𝑏 |(cos𝜃𝑖𝑗
𝑏𝑏 𝑏𝑏
+ 𝑠𝑖𝑛𝜃𝑖𝑗 ) (2.52)
𝜕𝛿𝑖𝑏
𝜕𝑄𝑖𝑐 𝑛
𝜕𝛿𝑖𝑐
= 𝑃𝑖𝑐 + ∑𝑗=2
𝑏𝑢𝑠
−|𝑌𝑖𝑗𝑐𝑐 ||𝑉𝑖𝑐 |(cos𝜃𝑖𝑗
𝑐𝑐 𝑐𝑐
+ 𝑠𝑖𝑛𝜃𝑖𝑗 ) (2.53)
𝜕𝑄𝑖𝑏 𝑄𝑖𝑏 𝑛
= 𝑏𝑢𝑠
+ ∑𝑗=2 −|𝑌𝑖𝑗𝑏𝑏 ||𝑉𝑖𝑏 |𝑠𝑖𝑛𝜃𝑖𝑗
𝑏𝑏
(2.54)
𝜕𝑉𝑖𝑏 |𝑉𝑖𝑏 |
𝜕𝑄𝑖𝑐 𝑄𝑖𝑐 𝑛
= 𝑏𝑢𝑠
+ ∑𝑗=2 −|𝑌𝑖𝑗𝑐𝑐 ||𝑉𝑖𝑐 |𝑠𝑖𝑛𝜃𝑖𝑗
𝑐𝑐
(2.55)
𝜕𝑉𝑖𝑐 |𝑉𝑖𝑐 |
149
Appendix B: Wind, PV and load models
In this work, the wind units (WTs) and photovoltaic units (PVs) are interfaced with the power
electronic converter and able to track a MPPT power for the generation. As such, the WTs and PVs
can be modeled as the sources of static injection power into the microgrid where their amount of
power generated are respected to the capacity windspeed and irradiance, respectively. In this way the
power generated via these resources can be calculated using the static model [78] and describe as
following:
0, 𝑣𝑡𝑚 < 𝑣 𝑐𝑖 , 𝑣𝑡𝑚 ≥ 𝑣 𝑐𝑜

𝑊𝑇,3𝛷 𝑊𝑇,𝑚𝑎𝑥
𝑃𝑖,𝑡 = {𝑆𝑖 𝑃𝐹𝑖𝑊𝑇 (𝑣𝑡𝑚 − 𝑣 𝑐𝑖 )⁄(𝑣 𝑟𝑎 − 𝑣 𝑐𝑖 ) , 𝑣 𝑐𝑖 ≤ 𝑣𝑡𝑚 ≤ 𝑣 𝑟𝑎
𝑆𝑖𝑊𝑇,𝑚𝑎𝑥 × 𝑃𝐹𝑖𝑊𝑇 , 𝑣 𝑟𝑎 < 𝑣𝑡𝑚 < 𝑣 𝑐𝑜
,∀𝑖 ∈ Ω𝑊𝑇 , ∀𝑡 ∈ 𝑁𝑇 (4.83)
then, the reactive power generated by the WTs can be calculated respecting to their specified setting
power factor. The formulation used for this calculation are illustrated as following:
𝑊𝑇,3𝛷 𝑊𝑇,3𝛷
𝑄𝑖,𝑡 = 𝑃𝑖,𝑡 𝑡𝑔(cos−1(𝑃𝐹𝑖𝑊𝑇 )) ,∀𝑖 ∈ Ω𝑊𝑇 , ∀𝑡 ∈ 𝑁𝑇 (4.84)
next, the PVs assumed controlled in the power factor unity mode in this work. As such, there is no
reactive power participated via the PV units. The static model used for the PV units can be:
𝑆𝑖𝑃𝑉,𝑚𝑎𝑥 (𝑟𝑡𝑚 )2 ⁄(𝑅 𝑆𝑇𝐷 − 𝑅 𝐶 ) , 𝑟𝑡𝑚 ≤ 𝑅 𝐶

𝑃𝑉,3𝛷
𝑃𝑖,𝑡 ={ 𝑆𝑖𝑃𝑉,𝑚𝑎𝑥 (𝑟𝑡𝑚 )2 ⁄𝑅 𝑆𝑇𝐷 , 𝑅 𝐶 < 𝑟𝑡𝑚 < 𝑅 𝑆𝑇𝐷 ,∀𝑖 ∈ Ω𝑃𝑉 , ∀𝑡 ∈ 𝑁𝑇 (4.85)
𝑆𝑖𝑃𝑉,𝑚𝑎𝑥 , 𝑟𝑡𝑚 ≥𝑅 𝑆𝑇𝐷
In term of the demand, the load are modeled by considering the pattern percentage of the annual peak
demands [90]. In this way, load at the time 𝑡 can be calculated though the multiplication of their peak
with the annual peak load factor along the planning horizon.
𝐿 𝐿,𝑝𝑒𝑎𝑘
𝑃𝑖,𝛹,𝑡 = (𝐴𝑃𝐿𝑡 ⁄100) × 𝑃𝑖,𝛹 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , 𝛹 ∈ ℱ, ∀𝑡 ∈ 𝑁𝑇 (4.86)
𝐿 𝐿,𝑝𝑒𝑎𝑘
𝑄𝑖,𝛹,𝑡 = (𝐴𝑃𝐿𝑡 ⁄100) × 𝑄𝑖,𝛹 ,∀𝑖 ∈ Ω𝐵𝑈𝑆 , 𝛹 ∈ ℱ, ∀𝑡 ∈ 𝑁𝑇 (4.87)
150
Appendix C: Algorithm & Example for MGs Formation
C.1 Algorithm for MGs Formation
𝒇𝒂𝒊𝒍
TABLE C. 1: Algorithm for determining 𝑭𝑨𝒃𝒖𝒔 {𝒃𝒓𝒊 }
Algorithm: formation of 𝑭𝑨𝒃𝒖𝒔

1: Initialize 𝑭𝑨𝒃𝒖𝒔 (1: 𝑛𝑏𝑢𝑠 ) = 1
𝑓𝑎𝑖𝑙
2: Generate 𝐏𝐀𝐓𝐇 𝐛𝐫 and 𝐏𝐀𝐓𝐇 𝐛𝐮𝐬 using path finder for 𝑏𝑟𝑖
3: for 𝑝𝑡ℎ = 1: 𝑛𝑝𝑎𝑡ℎ
𝑓𝑎𝑖𝑙 𝑓𝑎𝑖𝑙 𝑓𝑎𝑖𝑙
4: Set 𝑭𝑨𝒃𝒖𝒔 (𝑏𝑟𝑖 , [𝒇𝒃(𝑏𝑟𝑖 ), 𝒕𝒃(𝑏𝑟𝑖 )] = 0
5: for 𝑏𝑟 = 1: 𝑛𝑝𝑎𝑡ℎ
6: if 𝐏𝐀𝐓𝐇 𝐛𝐫 (𝑝𝑡ℎ, 𝑏𝑟) corresponding to location of ISW
7: Denote 𝑝𝑖𝑠 = 𝐏𝐀𝐓𝐇 𝐛𝐫 (𝑝𝑡ℎ, 𝑏𝑟)
8: Find 𝑎𝑓𝑏 is an element address of 𝐏𝐀𝐓𝐇 𝐛𝐫 (𝑝𝑡ℎ, 𝑏𝑟) = 𝒇𝒃(𝑝𝑖𝑠)
9: Find 𝑎𝑡𝑏 is an element address of 𝐏𝐀𝐓𝐇 𝐛𝐫 (𝑝𝑡ℎ, 𝑏𝑟) = 𝒇𝒃(𝑝𝑖𝑠)
10: 𝐅𝐀𝐛𝐮𝐬 (𝑏𝑟 𝑓𝑎𝑖𝑙 , 𝐏𝐀𝐓𝐇 𝐛𝐮𝐬 (𝑝𝑡ℎ, 1: 𝑚𝑖𝑛(𝑎𝑓𝑏 , 𝑎𝑡𝑏 )) = 0
11: Break for loop and go to next 𝑝𝑡ℎ
12: elseif there is no ISW along path 𝑝𝑡ℎ by 𝑏𝑟 = 𝑛𝑝𝑎𝑡ℎ
13: Find index for the non-zero number 𝒏𝒛 from 𝐏𝐀𝐓𝐇 𝐛𝐮𝐬 (𝑝𝑡ℎ, : )
14: 𝐅𝐀𝐛𝐮𝐬 (𝑏𝑟 𝑓𝑎𝑖𝑙 , 𝐏𝐀𝐓𝐇 𝐛𝐮𝐬 (𝑝𝑡ℎ, 𝒏𝒛)) = 0
15: Break for loop and go to next 𝑝𝑡ℎ
16: end if of elseif
17: end for
18:end for
TABLE C. 2: Formation of matrix containing microgrid set

Algorithm: formation of 𝐌𝐆𝑡𝐬𝐞𝐭
1: Initialize the matrix 𝐌𝐆𝑡𝐬𝐞𝐭 (1: 𝑛𝑏𝑢𝑠 , 1: 𝑛𝑏𝑢𝑠 ) = 0
𝐈−𝐈𝐈 (1:
2: Determine 𝒇𝒍𝒐𝒄𝑡 = 𝑓𝑖𝑛𝑑(𝐀𝐕𝑠,𝑠𝑒 𝑛𝑏𝑢𝑠 , 𝑡) = 0)
3: if 𝒇𝒍𝒐𝒄𝑡 is empty
4: Set the matrix 𝐌𝐆𝑡𝐬𝐞𝐭 (1: 𝑛𝑏𝑢𝑠 , 1: 𝑛𝑏𝑢𝑠 ) = 0
5: else
6: Import 𝑳𝑭𝒂𝑡 from 𝐋𝐅𝐚, then 𝑩𝑭𝒂𝑡 from B𝐅𝐚
7: Initialized the root of microgrid 𝑟𝑜𝑜𝑡 𝑀𝐺 = 0
8: Set and initialize 𝐌𝐆𝐬𝐞𝐭𝟎 = 0
9: Initialize the status of the isolate switch by 𝑠𝑡𝑛𝑖𝑠𝑤 = 0
➔ Find 𝑠𝑡𝑛𝑖𝑠𝑤 of the set 𝑰𝑺𝑾𝑠𝑖𝑡𝑒 for time t
10: for 𝑛 = 1: 𝑛𝑖𝑠𝑤
11: if 𝑳𝑭𝒂𝑡 (𝑰𝑺𝑾𝑛𝑠𝑖𝑡𝑒 ) = 0,
12: The 𝑰𝑺𝑾𝑛𝑠𝑖𝑡𝑒 is open and set 𝑠𝑡𝑛𝑖𝑠𝑤 = 0
13: else
151
14: The 𝑰𝑺𝑾𝑛𝑠𝑖𝑡𝑒 is still close and set 𝑠𝑡𝑛𝑖𝑠𝑤 = 1
15: end if
16: end for
17: Define 𝑰𝑺𝑾𝑜𝑝𝑒𝑛 = 𝑰𝑺𝑾𝒔𝒊𝒕𝒆 (𝑓𝑖𝑛𝑑(𝑠𝑡𝑛𝑖𝑠𝑤 = 0))
➔ Determine the adjacent matrix for 𝐌𝐆𝑡𝐬𝐞𝐭
18: Generate adjacency matrix 𝐀𝐌𝐆 for the initial test system using (105)
𝑜𝑝𝑒𝑛
19: for 𝑛 = 1: 𝑛𝑖𝑠𝑤
𝑜𝑝𝑒𝑛 𝑜𝑝𝑒𝑛
20: Set 𝐀𝐌𝐆 (𝒇𝒃(𝐼𝑆𝑊𝑛 ), 𝒕𝒃(𝐼𝑆𝑊𝑛 )) = 0
𝑜𝑝𝑒𝑛 𝑜𝑝𝑒𝑛
21: Set 𝐀𝐌𝐆 (𝒕𝒃(𝐼𝑆𝑊𝑛 ), 𝒇𝒃(𝐼𝑆𝑊𝑛 )) = 0
22: end for
➔ Determine the root of the microgrid(s) set
23: Determine 𝒃𝒖𝒔𝒐𝒏 , by 𝒃𝒖𝒔𝒐𝒏 = 𝑓𝑖𝑛𝑑(𝑩𝑭𝒂𝑡 = 1)
𝑜𝑝𝑒𝑛
24: for 𝑚 = 1: 𝑛𝑖𝑠𝑤
𝑜𝑝𝑒𝑛
25: if 𝐼𝑆𝑊𝑚 corresponding to the fault location
𝑜𝑝𝑒𝑛
26: Set 1 root by 𝑟𝑜𝑜𝑡 𝑀𝐺 (𝑚, 1) = 𝑓𝑏(𝐼𝑆𝑊𝑚 )
st
𝑜𝑝𝑒𝑛
27: Set 2nd root by 𝑟𝑜𝑜𝑡 𝑀𝐺 (𝑚, 2) = 𝑡𝑏(𝐼𝑆𝑊𝑚 )
28: else
𝑜𝑝𝑒𝑛
29: Determine a single root node 𝑟𝑜𝑜𝑡 𝑀𝐺 (𝑚, 1) equal to 𝑓𝑏 or 𝑡𝑏 of 𝐼𝑆𝑊𝑚
corresponding to the online node
30: end if
31: end for
32: Do the intersection between 𝐫𝐨𝐨𝐭 𝐌𝐆 and 𝒃𝒖𝒔𝒐𝒏 for eliminating the overlap and selecting
only the corresponding online node for forming vector 𝒓𝒐𝒐𝒕𝑴𝑮
➔ Forming 𝐌𝐆𝑡𝐬𝐞𝐭 based on Depth First Search (DFS)
33: for 𝑚 = 1: 𝑛𝑟𝑜𝑜𝑡
34: Do DFS on 𝑨𝐌𝐆 for 𝑟𝑜𝑜𝑡𝑚 𝑀𝐺
then store resulting vector in 𝑴𝑮𝑫𝑭𝑺
𝑚
35: Sort 𝑴𝑮𝑫𝑭𝑺𝑚
36: Update 𝐌𝐆𝐬𝐞𝐭𝟎 (𝑚, 1: 𝑙𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑀𝐺𝑚 𝐷𝐹𝑆 )
= 𝑴𝑮𝑫𝑭𝑺𝑚
37: end for
38: Check and eliminate the overlapping set of microgrid(s) in 𝐌𝐆𝐬𝐞𝐭𝟎
39: Store 𝐌𝐆𝐬𝐞𝐭𝟎 in 𝐌𝐆𝑡𝐬𝐞𝐭
40: end if
TABLE C. 3: Algorithm for determining 𝒃𝒎

𝒕
Algorithm: formation of 𝒃𝒎
𝑡
1: Initialize 𝒃𝒎𝑡 (1: 𝑛𝑏𝑢𝑠 ) = 1
2: if there is no fault(s) happening at time t
3: Set vector 𝒃𝒎𝑡 (1: 𝑛𝑏𝑢𝑠 ) to 0
4: else
5: Find the set microgrid in 𝐌𝐆𝑡𝐬𝐞𝐭 that is adjacent to the main grid through bus 1
6: Set 𝒃𝒎𝑡 to 0 for all buses corresponding to the set of microgrid that is adjacent to bus 1
7: end if
152
TABLE C. 4: Algorithm for forming 𝒃𝒖𝒔𝒏𝒆
𝒕
Algorithm: formation of 𝒃𝒖𝒔𝑛𝑒 𝑡

1: Initialize 𝒃𝒖𝒔𝒏𝒆𝒕 (1: 𝑛 𝑏𝑢𝑠 ) = 1
2: if there is no fault(s) happening at time t
3: All buses are energized by GC mode, set 𝒃𝒖𝒔𝒏𝒆 𝒕 (1: 𝑛𝑏𝑢𝑠 ) = 1
4: else
5: Define 𝑛𝑀𝐺 as the number of microgrid(s) from 𝐌𝐆𝑡𝐬𝐞𝐭
6: for 𝑚 = 1: 𝑛𝑀𝐺
7: if 𝑴𝑮𝒔𝒆𝒕
𝑚,𝑡 adjacent to the main grid
8: do nothing and jump to the next microgrid m of the for loop
9: elseif 𝑴𝑮𝒔𝒆𝒕
𝑚,𝑡 have a DDG unit installed by checking the element of
𝒔𝒆𝒕 𝑠𝑖𝑡𝑒 𝑠𝑖𝑧𝑒
𝑴𝑮𝑚,𝑡 with 𝑫𝑫𝑮𝑠𝑝 as well as confirm that 𝑫𝑫𝑮𝑠𝑝 ≠0
10: do nothing and jump to the next m of for loop
11: else
12: Define vector index 𝒊𝑛𝑒 that is equal to the non-zero element of 𝑴𝑮𝒔𝒆𝒕
𝑚,𝑡
13: Set 𝒃𝒖𝒔𝒏𝒆𝒕 (𝒊𝑛𝑒 )
= 0
14: end if
15:end for
TABLE C. 5: Algorithm for determining 𝒇𝒃𝒖𝒔

𝒕 and 𝑫𝑫𝑮𝒅𝒐𝒘𝒏
𝒕
Algorithm: formation of 𝒇𝒃𝒖𝒔

𝑡
1: Initialize vector 𝒇𝒃𝒖𝒔
𝑡 (1: 𝑛𝑑𝑑𝑔 ) = 1 and vector 𝑫𝑫𝑮𝑡
𝑑𝑜𝑤𝑛
(1: 𝑛𝑑𝑑𝑔 ) = 1
2: if there is no fault(s) happen at time t, it means MG operate in GC mode
3: Do nothing and let 𝒇𝒃𝒖𝒔 𝑡 (1: 𝑛𝑑𝑑𝑔 ) and 𝑫𝑫𝑮𝑡
𝑑𝑜𝑤𝑛
(1: 𝑛𝑑𝑑𝑔 ) as initial value
4: else
5: for 𝑛 = 1: 𝑛𝑑𝑑𝑔
5: Initialize 𝒉𝒅𝒅𝒈 (1: 𝑛𝑀𝐺 ) = 0
6: for 𝑚 = 1: 𝑛𝑀𝐺
7: if there is an element of 𝑴𝑮𝒔𝒆𝒕
𝑚,𝑡 corresponding to 𝐷𝐷𝐺𝑛,𝑠𝑝
𝑠𝑖𝑡𝑒
𝑑𝑑𝑔
8: Set ℎ𝑚 = 1
9: else
𝑑𝑑𝑔
10: Set ℎ𝑚 = 0
12: end if
13: end for
14: if ∑ 𝒉𝒅𝒅𝒈 = 0 means 𝐷𝐷𝐺𝑛,𝑠𝑝𝑠𝑖𝑡𝑒
are not stay on any MG set in 𝐌𝐆𝑡𝐬𝐞𝐭
𝑏𝑢𝑠 𝑠𝑖𝑡𝑒 𝑑𝑜𝑤𝑛
15: Set 𝑓𝑛,𝑡 = 𝐷𝐷𝐺𝑛,𝑠𝑝 and 𝐷𝐷𝐺𝑛,𝑡 =0
16: else
𝑏𝑢𝑠
17: Set 𝑓𝑛,𝑡 = 𝑴𝑮𝒔𝒆𝒕
𝑚,𝑡 (ℎ
𝑑𝑑𝑔
= 1,1) and 𝐷𝐷𝐺𝑛,𝑡 𝑑𝑜𝑤𝑛
=1
18: end if
19: end for
20: end if
153
C.2 Mathematical example for MGs Formation
The reference 8-bus test system with a fault located at branch #2 and #4 respective paths that can be
explored is described in Figure C. 1.
storm
MW #1 #2
MVAR
ISWsub 1 2 3
3
4 5
!
1 2 4
BUSBAR 7
#4
5
SUBSTATION #3 8
6
6 7
FIGURE C. 1: Path(s) search based on fault at branch #2
C.2.1 Fault influence path
a) Step 1: Representation of the System Topology
According to Figure C. 1, Figure C. 2 shows the undirected graph for the reference 8-bus system with
nodes (vertices) and branches (edges). In this case, the adjacency matrix is determined using (5.105),
and can be expressed for this reference system by (5.154).
8
2 3 4
Graph representation 3 4 5
Gbus(NDbus,EDbus) 1
2
1
5 6
6 7
FIGURE C. 2: Undirected Graph Tree for 8-bus test system
𝒃𝒖𝒔 # 1 2 3 4 5 6 7 8
# − − − − − − − −
1 | 0 1 0 0 0 0 0 0
2 | 1 0 1 0 0 1 0 0
3 | 0 1 0 1 0 0 0 0
𝐌𝐚𝐝𝐣 = (5.154)
4 | 0 0 1 0 1 0 0 1
5 | 0 0 0 1 0 0 0 0
6 | 0 1 0 0 0 0 1 0
7 | 0 0 0 0 0 1 0 0
[ 8 | 0 0 0 1 0 0 0 0]
154
b) Step 2: Determination of the Target Nodes 𝑻𝑮𝒏𝒐𝒅𝒆
This perspective is clearly described in (5.155) and Figure C. 3 for the reference 8-bus system to a
fault 𝑏𝑟 𝑓𝑎𝑖𝑙 = 2, where identified targets nodes are denoted by node #1 (root node), node #5 (leaf
𝑏𝑢𝑠 𝑛 𝑎𝑑𝑗
node), node #7 (leaf node) and node #8 (leaf node), e.g., for node #8, ∑𝑗=1 𝑀8,𝑗 = 1.
∑𝑛𝑗=1
𝑏𝑢𝑠 𝑎𝑑𝑗
𝑀𝑖,𝑗
−−−
𝟏
3
𝑻𝑮𝑛𝑜𝑑𝑒 = 𝑓𝑖𝑛𝑑 2 = 1 = [𝟏 𝟓 𝟕 𝟖] (5.155)
3
𝟏
2
𝟏
([ 𝟏 ] )
8
Start node 2 3 4
Leaf target nodes
1
3 4 5
Root target node 2
1
5 6
6 7
FIGURE C. 3: Start, Root and leaves nodes
c) Step 3: Determination of the Path(s)
The 𝐏𝐀𝐓𝐇 𝐛𝐮𝐬 and 𝐏𝐀𝐓𝐇 𝐛𝐫 matrices for the reference 8-bus system with fault at branch #2 can be
expressed by (5.156)-(5.157).
𝒑𝒂𝒕𝒉𝒃𝒖𝒔 𝑇𝐺𝑖𝑏𝑢𝑠 𝑓𝑏(𝑏𝑟 𝑓𝑎𝑖𝑙 ) → → →

# 𝑖= −− −− −− −−
1 1 | 2 1 0 0
𝐏𝐀𝐓𝐇 𝐛𝐮𝐬 = (5.156)
2 5 | 2 3 4 5
3 7 | 2 6 7 0
[ 4 8 | 2 3 4 8 ]
𝒑𝒂𝒕𝒉𝒃𝒓 𝑏𝑟 𝑓𝑎𝑖𝑙 → →
# −− −− −−
1 | 2 1 0
𝐏𝐀𝐓𝐇 𝐛𝐫 = (5.157)
2 | 2 3 4
3 | 2 5 6
[ 4 | 2 3 7 ]
155
C.2.2 Branch affected matrix (LFa)
a) Determination of 𝑭𝑨𝒃𝒓
𝑓𝑎𝑖𝑙
For determining 𝐅𝐀𝐛𝐫 , as shown in Figure 5. 8 the 𝐏𝐀𝐓𝐇 𝐛𝐫 is determined for 𝑏𝑟𝑖 ∈ Ω𝐿𝐼𝑁 . Then,
𝑓𝑎𝑖𝑙 𝑓𝑎𝑖𝑙
the row corresponding to 𝑏𝑟𝑖 for 𝐅𝐀𝐛𝐫 is obtained by searching 𝐏𝐀𝐓𝐇 𝐛𝐫 {𝑏𝑟𝑖 }. For example,
Figure C. 4 shows the reference 8-bus system equipped with four isolate switches 𝑰𝑺𝑾𝑠𝑖𝑡𝑒 =
𝑓𝑎𝑖𝑙
[1,4,5,7] for the fault 𝑏𝑟2 .
storm
MW #1 #2
MVAR
ISWsub 1 2 3
3
4 5
ISW1 ISW2
!
1 2 4
BUSBAR 7 ISW4
#4
ISW3 5
SUBSTATION #3 8
6
6 7
FIGURE C. 4: Simple 8-bus system with found ISWs and fault at branch #2
Then, for this case:
𝒑𝒂𝒕𝒉𝒃𝒓 𝑏𝑟 𝑓𝑎𝑖𝑙 → →
# −− −− −−
𝑓𝑎𝑖𝑙 1 | 𝟐 𝟏 0
𝐏𝐀𝐓𝐇 𝐛𝐫 (𝑏𝑟2 ) = (5.158)
2 | 𝟐 𝟑 𝟒
3 | 𝟐 𝟓 6
[ 4 | 𝟐 𝟑 𝟕 ]
First, 𝑭𝑨𝒃𝒓 is initialized by 𝐹𝐴𝑏𝑟 = [1 1 1 1 1 1 1 1]. Next, all branches that have been affected by
𝑓𝑎𝑖𝑙
the fault 𝑏𝑟2 , i.e., branch numbers highlighted in bold in (5.158), must have the fault cleared. In
this sense, for paths #1, #2, #3 and #4 the fault has been cleared by the opening the ISWs localized at
the branch #1, #4, #5 and #7; respectively. In contrast, the branch #6 has been protected by the ISW3,
thus not being highlighted. Finally, all elements of 𝑭𝑨𝒃𝒓 that correspond to affected branches, i.e.,
𝑓𝑎𝑖𝑙
[1,2,3,4,5,7], are set to zero. The vector 𝑭𝑨𝒃𝒓 corresponding to 𝑏𝑟2 and respecting to 𝑰𝑺𝑾𝑠𝑖𝑡𝑒 can
be expressed as follows:
𝑏𝑟𝑖 : 1 2 3 4 5 6 7
𝒃𝒓 𝑓𝑎𝑖𝑙 (5.159)
𝑭𝑨𝟐 (𝑏𝑟2 ) = [0 0 0 0 0 1 0]
156
Applying the same methodology while respecting assumptions 1) and 2) discussed in the beginning
of this section, the full matrix 𝐅𝐀𝐛𝐫 for the reference 8-bus test system can be formed as follows:
𝑏𝑟𝑖 1 2 3 4 5 6 7
𝑓𝑎𝑖𝑙
𝑏𝑟𝑖 − − − − − − −
1 | 0 1 1 1 1 1 1
2 | 0 0 0 0 0 1 0
𝐅𝐀𝐛𝐫 = 3 | 0 0 0 0 0 1 0 (5.160)
4 | 1 1 1 0 1 1 1
5 | 1 1 1 1 0 1 1
6 | 1 1 1 1 0 0 1
[ 7 | 1 1 1 1 1 1 0]
a) Branch affected matrix (LFa) considering single fault location
The binary vector 𝑳𝑭𝒂𝑡 can be obtained easily by searching a row of 𝐅𝐀𝐛𝐫 corresponding to the
location of the fault(s), i.e., 𝑓𝑙𝑜𝑐𝑖,𝑡 for scenario s season 𝑠𝑒. For the case shown in Figure C. 4, there
is only a single fault with 𝑓𝑙𝑜𝑐𝑡 = 2, in this case the vector 𝑳𝑭𝒂𝑡 can be given as follows:
𝑳𝑭𝒂𝒕 = 𝐅𝐀𝐛𝐫 (𝑓𝑙𝑜𝑐𝑡 , : ) = [0 0 0 0 0 1 0] (5.161)
a) Branch affected matrix (LFa) considering multiple fault location
To exemplify the case with multiple fault locations, assuming that faults simultaneously occur at
branch #3 and #6 in the reference 8-bus system, i.e., 𝒇𝒍𝒐𝒄𝑡 = [3, 6]. The vector 𝑳𝑭𝒂𝑡 can be obtained
using (5.162)-(5.164).
𝑳𝑭𝒂𝑡 {𝑓𝑙𝑜𝑐1,𝑡 = 3} = [0 0 0 0 0 1 0] (5.162)
𝑳𝑭𝒂𝑡 {𝑓𝑙𝑜𝑐1,𝑡 = 6} = [1 1 1 1 0 0 1] (5.163)
𝑳𝑭𝒂𝑡 {𝑓𝑙𝑜𝑐1,𝑡 = 3}
𝑳𝑭𝒂𝑡 (𝒇𝒍𝒐𝒄𝑡 ) = ( ) = [0 0 0 0 0 0 0] (5.164)
∘ 𝑳𝑭𝒂𝑡 {𝑓𝑙𝑜𝑐1,𝑡 = 6}
If there no fault(s) occur at time 𝑡, especially in the normal scenarios, i.e., Ω𝑠 , 𝑳𝑭𝒂𝑡 = [1 1 1 1 1 1 1].
C.2.4 Bus affected matrix (BFa)
a) Determination of 𝑭𝑨𝒃𝒖𝒔
Figure C. 5 shows the buses affected by the fault at branch #2.
157
storm
MW #1 #2
MVAR
ISWsub 1
ISW1
2 3
3
4
ISW2
5
!
1 2 4
BUSBAR 7 ISW4
#4
ISW3 5
SUBSTATION #3 8
6
Bus affected by fault
6 7
FIGURE C. 5: Buses affected by the original fault at branch #2
𝑓𝑎𝑖𝑙
By using the algorithm in Table C. 1 for all 𝑏𝑟𝑖 ∈ Ω𝐵𝑅 , the 𝐅𝐀𝐛𝐮𝐬 obtained for the reference 8-bus
system can be expressed as follows:
𝑏𝑢𝑠𝑖 1 2 3 4 5 6 7 8
𝑓𝑎𝑖𝑙
𝑏𝑟𝑖 − − − − − − − −
1 | 1 1 1 1 1 1 1 1
2 | 1 0 0 0 1 1 1 1
𝐅𝐀𝐛𝐮𝐬 = 3 | 1 0 0 0 1 1 1 1 (5.165)
4 | 1 1 1 1 1 1 1 1
5 | 1 1 1 1 1 1 1 1
6 | 1 1 1 1 1 0 0 1
[ 7 | 1 1 1 1 1 1 1 1]
𝑓𝑎𝑖𝑙
As can be seen, comparing Figure C. 5, bus #2, #3 and #4 were deactivated by the fault 𝑏𝑟2 . In this
𝑓𝑎𝑖𝑙
case, 𝐅𝐀𝐛𝐮𝐬 (𝑏𝑟2 ,[2,3,4]) is set to zero.
a) Bus affected matrix (BFa) considering single fault location
The 𝑩𝑭𝒂𝑡 can be obtained by searching a row of 𝐅𝐀𝐛𝐮𝐬 corresponding to the location of the fault(s),
i.e., 𝑓𝑙𝑜𝑐𝑖,𝑡 , for scenario s of season se. For the case shows in Figure C. 5, there is only a single fault
𝑓𝑙𝑜𝑐𝑡 = 2 and the vector 𝑩𝑭𝒂𝑡 is obtained as follows:
𝑳𝑭𝒂𝒕 = 𝐅𝐀𝐛𝐫 (𝑓𝑙𝑜𝑐𝑡 , : ) = [1 0 0 0 1 1 1 1] (5.166)
a) Bus affected matrix (BFa) considering multiple fault location
To exemplify the case with multiple fault location, the reference 8-bus system with faults at the branch
#3 and #6 is assumed, i.e., 𝒇𝒍𝒐𝒄𝑡 = [3, 6]. In this case, the vector 𝑩𝑭𝒂𝑡 can be determined as:
𝑩𝑭𝒂𝑡 {𝑓𝑙𝑜𝑐1,𝑡 = 3} = [1 0 0 0 1 1 1 1] (5.167)
158
𝑩𝑭𝒂𝑡 {𝑓𝑙𝑜𝑐1,𝑡 = 6} = [1 1 1 1 1 0 0 1] (5.168)
𝑩𝑭𝒂𝑡 {𝑓𝑙𝑜𝑐1,𝑡 = 3}
𝑩𝑭𝒂𝑡 (𝒇𝒍𝒐𝒄𝑡 ) = ( ) = [1 0 0 0 1 0 0 1] (5.169)
∘ 𝑩𝑭𝒂𝑡 {𝑓𝑙𝑜𝑐1,𝑡 = 6}
For the case where there is no fault at time 𝑡, 𝑩𝑭𝒂𝑡 can be simply set to 𝑩𝑭𝒂𝑡 = [1 1 1 1 1 1 1 1].
C.2.5 Microgrid Formation set
Figure C. 6 describes the matrix 𝐌𝐆𝑡𝐬𝐞𝐭 including all possible microgrid(s) formation options for the
reference 8-bus system with fault located at branch #2.
storm
MW
MVAR
MG #1 MG #2
ISWsub 1
ISW1
2 3
3
4
ISW2
5
!
1 2 4
7 ISW4
ISW3 5 8
SUBSTATION
MG #4
6 7
MG #3
FIGURE C. 6: 𝐌𝐆𝐬𝐞𝐭 formation
In this case, 𝐌𝐆𝑡𝐬𝐞𝐭 can be expressed by (5.170):
𝑴𝑮 # 𝑏𝑢𝑠 # 𝑏𝑢𝑠 #
1 | 1 0
𝐌𝐆𝑡𝐬𝐞𝐭 = 2 | 5 0 (5.170)
3 | 6 7
[ 4 | 8 0 ]
C.2.6 Non-Energized system buses
Figure C. 7 exemplifies the microgrid formation for cases with and without local DDG installed
capacity. In Figure C. 7(a), the bus #1 was protected by ISW #1 and stays connected to the main grid.
According to ISW #2, #3 and #4, the MG #1, #2 and #3 can be formed, respectively. In term of
resources, the DDG unit located at bus #4 corresponds to the area affected by the fault at branch #2.
On the other hand, there are PV, WT and ESS located within the possibly formed microgrids #1, #2
and #3. Nevertheless, as the PV, WT and ESS could not operate with droop/2nd for maintaining the
voltage and system frequency, these microgrids are considered as non-successful microgrids for
159
operation in IS mode. Thus, except for bus #1 all the element of 𝒃𝒖𝒔𝑛𝑒
𝑡 are set to zero. Similar to case
(a), in Figure C. 7(b) three microgrids are formed. Among them, MG #3 could not operate in IS mode
as there is no DDG located within this area. However, the DDGs in this case were placed in the MG
#1 and #2, allowing these microgrids to operate in IS mode. In this case, bus #1 and the elements of
𝑡 corresponding to buses located within microgrids with successful IS mode can be set to one
(i.e., 𝒃𝒖𝒔𝑛𝑒
𝑡 = [1 0 0 0 1 1 1 0]).
storm
MW MG #1
MVAR
ISWsub 1
ISW1
2 3
3
4
ISW2
5
!
1 2 4
7 ISW4
ISW3 5 8
SUBSTATION
MG #3
6 7
MG #2
(a)
MG #1
storm
MW
MVAR
ISWsub 1
ISW1
2 3
3
4
ISW2
5
!
1 2 4
7 ISW4
ISW3 5 8
SUBSTATION
MG #3
6 7
MG #2
(b)
MG with successful MG without successful
Bus no-energize
operate in IS mode operated in IS mode
FIGURE C. 7: Microgrids formation and bus no-energize respecting to the location of DDGs
160
Appendix D: Feasible operating regions of DDGs
The operation range of the DDGs shown in Figure D. 1 has been derived with respect to the feasible
operating regions of DDGs [19].
Non-feasible operation region Feasible operation region

DDG
Q QDDG
PiDDG,min= 0 PiDDG,max
QiDDG,max
PDDG PDDG
QiDDG,min
Not permissible Not permissible

(a) (b)
QDDG QDDG
Not permissible PFiDDG,1st
SiDDG,max
PDDG PDDG
PFiDDG,4th Not permissible

(c) (d)
Combination operation boundary constraints

(a), (b), (c) and (d)
Boundary derivered by (a) Boundary derivered by (c)

Boundary derivered by (b) Boundary derivered by (d)
QDDG
DDG
Feasible operation region
PDDG
DDG
Non-feasible operation region
(e)
FIGURE D. 1: Derivation of DDG feasible operation region
161
In this work, it is assumed the DDGs are generating units and allowed to generate active power and
to exchange reactive power using the interfaced power electronic devices. Figure D. 1 shows the
procedure executed for extracting the DDGs’ feasible operation region. Table D. 1 summarizes the
relation between the constraints, Figure D. 1, and their relative parameters for determining the DDG’s
feasible operation region.
TABLE D. 1: Constraints and parameters related to the determination of DDG's feasible

operation region
Constraint Fig. Parameters
# D.1 𝑃𝑖𝐷𝐷𝐺,𝑚𝑖𝑛 𝑃𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 𝑄𝑖𝐷𝐷𝐺,𝑚𝑖𝑛 𝑄𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 𝑃𝐹𝑖𝐷𝐷𝐺,1𝑠𝑡 𝑃𝐹𝑖𝐷𝐷𝐺,4𝑡ℎ 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥
(4.26) (a) 0 [0, ℝ] - - - - -
(4.27) (b) - - [−ℝ, 0] [0, ℝ] - - -
(4.28) (c) - - - - [0, ℝ] [0, ℝ] -
(4.29) (d) - - - - - - [0, ℝ]
The DDG 𝑖 active power 𝑃𝑖𝐷𝐷𝐺,𝑚𝑖𝑛 in (4.26) is set to as low as zero as the DDGs are supposed to be
operating as a source of active power. As such, the left-hand side of the operating region is excluded
as shown in Figure D. 1(a)-(d). The 𝑃𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 in (4.26), on the other hand, can be set according to
the DDGs maximum allowable power generation to adjust the right-hand side operating region as
illustrated by Figure D. 1(a). Nonetheless, 𝑃𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 is usually set to 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 for allowing the
DDGs to generate their active power proportional to their size. Figure D. 1(b) illustrates the impact
of the reactive power limits, 𝑄𝑖𝐷𝐷𝐺,𝑚𝑖𝑛 and 𝑄𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 given by (4.27), on the DDGs feasible operation
region. They restrict the DDGs’ reactive power generated to and received from the main grid,
respectively. Hence, the DDGs feasible operating region is restricted. The power factor 𝑃𝐹𝑖𝐷𝐷𝐺,1𝑠𝑡
and 𝑃𝐹𝑖𝐷𝐷𝐺,4𝑡ℎ in (4.28) limit the DDG’s minimum generated active power proportional to its
generated and received reactive power, respectively. As shown in Figure D. 1(c), 𝑃𝐹𝑖𝐷𝐷𝐺,1𝑠𝑡 and
𝑃𝐹𝑖𝐷𝐷𝐺,4𝑡ℎ adjust the angle of the radius located in the first and fourth quadrant of the DDG’s
operating circle. Finally, Figure D. 1(d) demonstrates the impact of the quadratic power constraint
(4.29). As can be seen, increasing 𝑆𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 expands the perimeter of the DDGs feasible operation
circle. Finally, the feasible operation region for the DDGs can be determined by the constraints (4.26)-
(4.29) as shown in Figure D. 1(e).
162
Appendix E: Linearization of absolute value operator
E.1 Linearization of Absolute Value Operator
In order to linearize the nonlinear expression 𝑧 = |𝑦| in the problem with objective function Min |𝑦|,
where 𝑦 is a continuous variable and the | . | is the absolute value operator, the auxiliary constraints
(4.89)-(4.90) are considered and the problem is reformulated as follows [118]:
Min 𝑧 (4.88)
𝑦≤𝑧 (4.89)
−𝑦 ≤ 𝑧 (4.90)
where, (4.88)-(4.89) perform the function of the absolute value operator for 𝑦 ≥ 0, while (4.88) and
(4.90) do the same for 𝑦 ≤ 0.
E.2 Linearize the production between continuous and binary variables

Let 𝑧 = 𝐴 × 𝑏 be the expression of the production between the continuous variable (𝐴) and the binary
variable (𝑏), if 𝐴 is bounded between zero and the big M i.e., 𝐴𝑏𝑖𝑔𝑀 , then, a representing variable 𝑧
can be introduced into the problem formulation subjected to (4.91)-(4.94) [83]:
z ≤ 𝐴𝑏𝑖𝑔𝑀 × 𝑏 (4.91)
z≤𝐴 (4.92)
z ≥ 𝐴 − (1 − 𝑏)𝐴𝑏𝑖𝑔𝑀 (4.93)
0 ≤ z ≤ 𝐴𝑏𝑖𝑔𝑀 (4.94)
163
Appendix F: Decision variables for OPF/LOPF problem
The decision variables of the model are formed by the combination of the groups of vectors of control
and dependent variables which are illustrated in the table below. Here it is worth noting that the
decision vector is the same for both IS and GC operation modes. However, the relevant decision
variables, within the decision vector, are activated/disactivated by using the binary input parameter
𝑏𝑡𝑚 in the proposed problem.
TABLE F. 1: Control and dependent variables of the MINLP and MILP model for minimizing
their objective functions
Group Variables Type Set 𝑿𝑴𝑰𝑳𝑷,𝟑𝜱 𝑿𝑴𝑰𝑵𝑳𝑷,𝟑𝜱 Operation mode
# ∀𝑖, 𝑘 ∈ IS GC
1 𝑉𝑖,𝛹,𝑡 ,𝛹 ∈ ℱ ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐵𝑈𝑆 ✓ ✓ ✓ ✓
2 𝛿𝑖,𝛹,𝑡 ,𝛹 ∈ ℱ ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐵𝑈𝑆 ✓ ✓ ✓ ✓
𝑐𝑢𝑟𝑡
3 𝑏𝑖,𝑡 ,∀𝑡 ∈ 𝛺𝑇 ℤ Ω𝐵𝑈𝑆 ✓ ✓ ✓ -
𝑐𝑢𝑟𝑡
4 𝑧𝑖,𝑡 ,∀𝑡 ∈ 𝛺𝑇 ℤ Ω𝐵𝑈𝑆 ✓ - ✓ -
𝐷𝐷𝐺,3𝛷
5 𝑃𝑖,𝑡 ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐷𝐷𝐺 ✓ ✓ ✓ ✓
𝐷𝐷𝐺
6 𝑃𝑖,𝛹,𝑡 ,𝛹 ∈ ℱ ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐷𝐷𝐺 ✓ ✓ ✓ ✓
𝐷𝐷𝐺,3𝛷
7 𝑄𝑖,𝑡 ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐷𝐷𝐺 ✓ ✓ ✓ ✓
𝐷𝐷𝐺
8 𝑄𝑖,𝛹,𝑡 ,𝛹 ∈ ℱ ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐷𝐷𝐺 ✓ ✓ ✓ ✓
𝑣,𝑖𝑛𝑡
9 𝑢𝑖,𝛹,𝑡 ,𝛹 ∈ ℱ ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐷𝐷𝐺 ✓ ✓ ✓ -
𝑑𝑐ℎ,3𝛷
10 𝑃𝑖,𝑡 ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐸𝑆𝑆 ✓ ✓ ✓ ✓
𝑐ℎ,3𝛷
11 𝑃𝑖,𝑡 ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐸𝑆𝑆 ✓ ✓ ✓ ✓
𝐸𝑆𝑆,3𝛷
12 𝑄𝑖,𝑡 ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐸𝑆𝑆 ✓ ✓ ✓ ✓
𝐸𝑆𝑆
13 𝐸𝑖,𝑡 ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐸𝑆𝑆 ✓ ✓ ✓ ✓
14 𝑏𝐸𝑆𝑆
𝑖,𝑡 ,∀𝑡 ∈ 𝛺𝑇 ℤ Ω𝐸𝑆𝑆 ✓ ✓ ✓ ✓
𝑃𝐶𝐶,3𝛷
15 𝑃𝑖,𝑡 ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝑃𝐶𝐶 ✓ ✓ - ✓
𝑃𝐶𝐶
16 𝑃𝑖,𝛹,𝑡 ,𝛹 ∈ ℱ ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝑃𝐶𝐶 ✓ ✓ - ✓
𝑃𝐶𝐶,3𝛷
17 𝑄𝑖,𝑡 ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝑃𝐶𝐶 ✓ ✓ - ✓
𝑃𝐶𝐶
18 𝑄𝑖,𝛹,𝑡 ,𝛹 ∈ ℱ ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝑃𝐶𝐶 ✓ ✓ - ✓
𝑏𝑟
19 𝑃𝑘,𝛹,𝑡 ,𝛹 ∈ ℱ ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐿𝐼𝑁 ✓ ✓ ✓ ✓
𝑏𝑟
20 𝑄𝑘,𝛹,𝑡 ,𝛹 ∈ ℱ ,∀𝑡 ∈ 𝛺𝑇 ℝ Ω𝐿𝐼𝑁 ✓ ✓ ✓ ✓
164
Appendix G: Data information
G.1 Generator’s information
TABLE G. 1: DGs locations and static droop coefficients
Test system DG # Bus # 𝒎𝒑 𝒏𝒒 𝒇∗ 𝑽∗
(p.u.) (p.u.) (p.u.) (p.u.)
33-bus 1 1 0.05 0.05 1 1
Balanced 2 6 1.00 1.00 1 1
system 3 13 0.10 0.10 1 1
4 25 1.00 1.00 1 1
5 33 0.20 0.20 1 1
10476-bus 1 5 0.000042 0.00069 1 1.01
Balanced 2 18 0.000057 0.00095 1 1.01
system 3 120 0.000057 0.00095 1 1.01
4 345 0.000063 0.00104 1 1.01
5 598 0.000057 0.00095 1 1.01
6 848 0.000052 0.00087 1 1.01
7 5320 0.000078 0.00130 1 1.01
8 6913 0.000096 0.00160 1 1.01
9 8506 0.000089 0.00149 1 1.01
10 9568 0.000069 0.00116 1 1.01
11 8508 0.000078 0.00130 1 1.01
12 5851 0.000066 0.00110 1 1.01
13 6382 0.000114 0.00189 1 1.01
14 6913 0.000057 0.00095 1 1.01
15 7444 0.000057 0.00095 1 1.01
16 7975 0.000052 0.00087 1 1.01
17 9595 0.000066 0.00110 1 1.01
18 9037 0.000069 0.00116 1 1.01
19 9568 0.000078 0.00130 1 1.01
20 10012 0.000052 0.00087 1 1.01
25-bus 1 13 0.005 0.05 1 1.01
Unbalanced 2 19 0.01 0.10 1 1.01
system 3 25 0.005 0.05 1 1.01
165
G.2 Data information exclusive for Chapter 4
TABLE G. 2: Daily information data input for the 33-bus and 25-bus test system
Time 𝝈𝑷𝑪𝑪
𝒕 𝑨𝑷𝑳𝒕 𝒘𝒕 𝒓𝒕 Time 𝝈𝑷𝑪𝑪
𝒕 𝑨𝑷𝑳𝒕 𝒘𝒕 𝒓𝒕
(h) ($/MWh) (%) (m/s) (kW/m2) (h) ($/MWh) (%) (m/s) (kW/m2)
1:00 15.63 64.0 8.1 0 13:00 15.33 98.0 7.1 0.916
2:00 15.50 60.0 7.6 0 14:00 16.21 100.0 4.8 0.206
3:00 14.63 58.0 7.6 0 15:00 17.21 100.0 3.8 0.390
4:00 13.28 55.0 6.9 0 16:00 17.26 96.0 3.3 0.117
5:00 12.88 55.0 6.9 0 17:00 18.48 95.0 2.2 0.629
6:00 12.31 57.0 7.0 0.391 18:00 20.03 95.0 0.5 0.368
7:00 12.07 64.0 7.0 0.212 19:00 21.85 93.0 2.2 0
8:00 11.85 76.0 7.3 0.371 20:00 20.92 92.0 2.6 0
9:00 12.10 88.0 6.2 0.592 21:00 20.06 92.0 4.8 0
10:00 12.87 95.0 7.2 0.752 22:00 19.17 93.0 6.2 0
11:00 14.01 98.0 7.1 0.868 23:00 17.41 88.0 4.8 0
12:00 14.88 100.0 7.5 0.893 24:00 16.15 72.0 5.1 0
TABLE G. 3: Load weight and maximum duration of load curtailment 33-bus and 25-bus test
system
33-bus (balanced) 25-bus (unbalanced)
Bus 𝒘𝒍𝒐𝒂𝒅
𝒊 𝑻𝒄𝒖𝒓𝒕
𝒊 Bus 𝒘𝒍𝒐𝒂𝒅𝒊 𝑻𝒄𝒖𝒓𝒕
𝒊 Bus 𝒘𝒍𝒐𝒂𝒅
𝒊 Bus 𝒘𝒍𝒐𝒂𝒅
𝒊
(h) (h) (h) (h)
1 0 24 18 0.024 19 1 0.000 24 18 0.037 17
2 0.027 19 19 0.024 19 2 0.000 24 19 0.056 15
3 0.024 19 20 0.024 19 3 0.033 17 20 0.033 17
4 0.032 18 21 0.024 19 4 0.047 15 21 0.037 17
5 0.016 21 22 0.024 19 5 0.037 17 22 0.047 15
6 0.016 21 23 0.024 19 6 0.037 17 23 0.056 15
7 0.054 14 24 0.113 4 7 0.000 24 24 0.033 17
8 0.054 14 25 0.113 4 8 0.037 17 25 0.056 15
9 0.016 21 26 0.016 21 9 0.056 15
10 0.016 21 27 0.016 21 10 0.033 17
11 0.012 21 28 0.016 21 11 0.042 17
12 0.016 21 29 0.032 18 12 0.047 15
13 0.016 21 30 0.054 14 13 0.033 17
14 0.032 18 31 0.04 16 14 0.047 15
15 0.016 21 32 0.057 14 15 0.124 4
16 0.016 21 33 0.016 21 16 0.037 17
17 0.016 21 17 0.037 17
166
TABLE G. 4: Characteristic of the DDG, WTs, PVs, ESS, transformer, and transmission line
33-bus (balanced)
𝒎𝒂𝒙
No. Bus 𝑺 𝑪𝑹𝑫𝑫𝑮
𝒊 Mode PF 𝑬𝒎𝒂𝒙 /𝑬𝒎𝒊𝒏 𝑰𝒎𝒂𝒙
(kVA) (cf/kWh) (GC/IS) (kWh) (A Ψ)
nd
DDGs 1 2 1500 11.105 PQ/2 - - -
2 9 900 7.806 PQ/2nd - - -
nd
3 8 1700 11.165 PQ/2 - - -
4 22 500 7.316 PQ/2nd - - -
nd
5 25 2000 11.418 PQ/2 - - -
WTs 1 18 100 - PQunity 1.00 - -
unity
2 24 150 - PQ 1.00 - -
3 27 100 - PQlag 0.95 - -
lag
4 32 100 - PQ 0.95 - -
PVs 1 24 100 - PQunity 1.00 - -
unity
2 29 200 - PQ 1.00 - -
3 31 200 - PQunity 1.00 - -
ESSs 1 21 350 - PQ - 550/160 -
2 25 200 - PQ - 400/80 -
3 33 200 - PQ - 400/80 -
TR - 1 5300 - - - - -
Line 1-8 - - - - - - 500
9-32 - - - - - - 200
25-bus (unbalanced)
𝒎𝒂𝒙
No. Bus 𝑺 𝑪𝑹𝑫𝑫𝑮
𝒊 Mode PF 𝑬𝒎𝒂𝒙 /𝑬𝒎𝒊𝒏 𝑰𝒎𝒂𝒙
(kVA) (cf/kW) (GC/IS) (kWh) (A Ψ)
nd
DDGs 1 13 2000 11.148 PQ/2 - - -
2 19 1400 11.105 PQ/2nd - - -
nd
3 25 2500 11.165 PQ/2 - - -
WTs 1 12 100 - PQunity 1.00 - -
lag
2 22 200 - PQ 0.95 - -
PVs 1 5 150 - PQunity 1.00 - -
unity
2 15 150 - PQ 1.00 - -
ESSs 1 5 250 - PQ - 500/100
2 17 350 - PQ - 700/140
TR - - 5000 - - - - -
Line 1-6 - - - - - - 700
7-24 - - - - - - 350
167
TABLE G. 5: General parameter information
Parameter Value
𝑃𝑖𝐷𝐷𝐺,𝑚𝑖𝑛 /𝑃𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 0/S max
𝑄𝑖𝐷𝐷𝐺,𝑚𝑖𝑛 /𝑄𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 −0.2Smax /+0.6Smax
∗ ∗∗
𝑉 /𝑉 1.01/1.05
𝑓𝑏𝑢𝑠 /𝑘𝑖 𝜔 2/4
𝐷𝐷𝐺,1𝑠𝑡
𝑃𝐹𝑖 /𝑃𝐹𝑖𝐷𝐷𝐺,4𝑡ℎ 0.4/0.4
𝜎 𝑓𝑢𝑒𝑙 0.007 $/Mcf [119]
𝐸𝑆𝑆,𝑚𝑖𝑛
𝑄𝑖 /𝑄𝑖𝐸𝑆𝑆,𝑚𝑎𝑥 −0.6Smax /+0.6Smax
𝜂𝑖𝐸𝑆𝑆 95 %
𝑇𝑅,1𝑠𝑡
𝑃𝐹𝑖 /𝑃𝐹𝑖𝑇𝑅,4𝑡ℎ 0.75/0.75
168
G.3 Data information exclusive for Chapter 5
TABLE G. 6: Set of scenarios and their probability obtained by fast forward selection and
their corresponding calendar day/year for WT and PV
Season Details
Fall Scenario {80, 352, 694, 717, 718}
Day # 324, 322, 300, 323, 324
Year 2008, 2011, 2015, 2015, 2015
𝒔,𝒔𝒆 {0.385, 0.024, 0.519, 0.048, 0.024}
Winter Scenario {291, 554, 586, 669, 854}
Day # 352, 342, 374, 366, 369
Year 2011, 2014, 2014, 2015, 2017
𝒔,𝒔𝒆 {0.042, 0.403, 0.120. 0.093. 0.341}
Spring Scenario {34, 103, 127, 563, 564}
Day # 94, 71, 95, 76, 77
Year 2008, 2009, 2009, 2014, 2014
𝒔,𝒔𝒆 {0.111, 0.021, 0.447, 0.405, 0.016}
Summer Scenario {152, 341, 342, 525, 619}
Day # 212, 219, 220, 221, 224
Year 2009, 2011, 2011, 2013, 2014
𝒔,𝒔𝒆 {0.523, 0.335, 0.014, 0.067, 0.061}
TABLE G. 7: The most vulnerable wind scenarios and their corresponding date
Season Corresponding date max (𝒗𝒘𝒐𝒓𝒔𝒕
𝒔𝒆 )
day month year m/s (m/s)gust
Fall 19 November 2012 15.83 20.37
Winter 18 January 2011 16.11 20.74
Spring 12 Mars 2012 16.66 21.44
Summer 01 June 2010 15.55 20.01
TABLE G. 8: Parameter for generating scenario

Parameters Unit Set belong Values
𝑝𝑟𝑒
𝑁𝑠 - - 10000 𝜇𝑖𝑝𝑜𝑙𝑒 /𝜎𝑖𝑝𝑜𝑙𝑒 - ∀𝑖 ∈ Ω𝐿𝐼𝑁 65/15
𝑁𝑠′ /𝑁𝑠′′ - - 5/5 𝜇𝑖𝑐𝑜𝑛𝑑 /𝜎𝑖𝑐𝑜𝑛𝑑 - ∀𝑖 ∈ Ω𝐿𝐼𝑁 47/9
𝑣𝑖𝑐𝑖 m/s ∀𝑖 ∈ Ω𝑐𝑎𝑛
𝑊𝑇 4 𝜋 𝑏𝑟,𝑛𝑣𝑢𝑙 - - 0.05
𝑣𝑖𝑐𝑜 m/s ∀𝑖 ∈ Ω𝑐𝑎𝑛
𝑊𝑇 26 𝑙𝑜𝑤 ℎ𝑖𝑔ℎ
𝑣𝑐𝑟𝑖 /𝑣𝑐𝑟𝑖 m/s - 20/40
𝑣𝑖𝑟𝑎 m/s 𝑐𝑎𝑛
∀𝑖 ∈ Ω𝑊𝑇 16 𝑀𝑇𝑇𝑅 𝑛𝑤 h - 2
𝑅𝑖𝐶 kW/m2 ∀𝑖 ∈ Ω𝑐𝑎𝑛
𝑃𝑉 0.15 ℎ1𝑙𝑜𝑤 /ℎ2𝑙𝑜𝑤 h - 2/3
𝑅𝑖𝑆𝑇𝐷 kW/m2 ∀𝑖 ∈ Ω𝑐𝑎𝑛
𝑃𝑉 1 ℎ𝑖𝑔ℎ ℎ𝑖𝑔ℎ
ℎ1 /ℎ2 h - 3/4
𝑔𝑢𝑠𝑡
𝑘 - - 1.287
169
TABLE G. 9: Characteristic of the 33-bus test system
𝑳,𝒑𝒆𝒂𝒌 𝑳,𝒑𝒆𝒂𝒌
Line fb tb R X 𝑳𝒃𝒓 𝑵𝒑𝒐𝒍𝒆
𝒊 𝑷𝒕𝒃 𝑸𝒕𝒃 𝒘𝒍𝒐𝒂𝒅
No. # # (ohm) (ohm) (m) (kW) (kVar)
1 1 2 0.092 0.047 100 4 100 60 0.027
2 2 3 0.492 0.251 500 14 90 40 0.024
3 3 4 0.365 0.186 350 10 120 80 0.032
4 4 5 0.380 0.194 350 10 60 30 0.016
5 5 6 0.817 0.706 800 21 60 20 0.016
6 6 7 0.187 0.618 200 6 200 100 0.054
7 7 8 0.710 0.235 700 19 200 100 0.054
8 8 9 1.028 0.739 1000 26 60 20 0.016
9 9 10 1.042 0.739 1000 26 60 20 0.016
10 10 11 0.196 0.065 200 6 45 30 0.012
11 11 12 0.374 0.124 350 10 60 35 0.016
12 12 13 1.465 1.153 1500 39 60 35 0.016
13 13 14 0.540 0.711 550 15 120 80 0.032
14 14 15 0.590 0.525 600 16 60 10 0.016
15 15 16 0.745 0.544 750 20 60 20 0.016
16 16 17 1.286 1.718 1300 34 60 20 0.016
17 17 18 0.731 0.573 700 19 90 40 0.024
18 2 19 0.164 0.156 150 5 90 40 0.024
19 19 20 1.501 1.353 1500 39 90 40 0.024
20 20 21 0.409 0.477 400 11 90 40 0.024
21 21 22 0.707 0.935 700 19 90 40 0.024
22 3 23 0.450 0.308 450 12 90 50 0.024
23 23 24 0.896 0.708 900 24 420 200 0.113
24 24 25 0.894 0.700 900 24 420 200 0.113
25 6 26 0.203 0.103 200 6 60 25 0.016
26 26 27 0.284 0.144 300 9 60 25 0.016
27 27 28 1.057 0.932 1000 26 60 20 0.016
28 28 29 0.802 0.699 800 21 120 70 0.032
29 29 30 0.506 0.258 500 14 200 600 0.054
30 30 31 0.972 0.961 950 25 150 70 0.040
31 31 32 0.310 0.361 300 9 210 100 0.057
32 32 33 0.340 0.529 350 10 60 40 0.016
Total 3715 2300
170
TABLE G. 10: General parameters
Elements Parameters Unit/detail Values Level problem
outer inner
DDG 𝜎 𝑓𝑢𝑒𝑙 $/kWh 0.0205 ✓ ✓
𝜂𝑖𝐷𝐷𝐺 per unit 0.370 ✓ ✓
𝐷𝐷𝐺
𝜎𝐼𝑁𝑉 $/kW 1300 ✓ ✓
𝐷𝐷𝐺
𝜎𝑂&𝑀 $/kWh 0.003 ✓ ✓
𝐷𝐷𝐺
𝐾𝐶𝑂2 kg/kWh 201.6 ✓ -
Mode for GC/IS PQ/2nd - ✓
𝑃𝑖𝐷𝐷𝐺,𝑚𝑎𝑥 MW 4.0 ✓ -
𝐷𝐷𝐺
𝑃𝑠𝑡𝑒𝑝 kW 100 ✓ -
𝑉 ∗ /𝑉 ∗∗ p.u. 1.01/1.05 - ✓
𝑘𝑖 𝜔 - 4 - ✓
𝑃𝐹𝑖𝐷𝐷𝐺,1𝑠𝑡 - 0.4 - ✓
𝑃𝐹𝑖𝐷𝐷𝐺,4𝑡ℎ - 0.4 - ✓
𝜒 𝐷𝐷𝐺,𝑚𝑖𝑛 /𝜒 𝐷𝐷𝐺,𝑚𝑎𝑥 - -0.2/0.6 - ✓
WT 𝑊𝑇
𝜎𝐼𝑁𝑉 $/kW 1686 ✓ -
𝑊𝑇
𝜎𝑂&𝑀 $/kW/year 46 ✓ -
Mode for GC/IS PQ unity
- ✓
𝑃𝑖𝑊𝑇,𝑚𝑎𝑥 MW 0.6 ✓ -
𝑊𝑇
PV 𝑃𝑉
𝜎𝐼𝑁𝑉 $/kW 2277 ✓ -
𝑃𝑉
Mode for GC/IS PQ unity
- ✓
𝑃𝑖𝑃𝑉,𝑚𝑎𝑥 MW 0.3 ✓ -
𝑃𝑉
ES 𝐸𝑆,𝑃
𝜎𝐼𝑁𝑉 $/kW 300 ✓ -
𝐸𝑆
𝐸𝑆,𝐸
𝜎𝐼𝑁𝑉 $/kWh 350 ✓ -
𝐸𝑆
𝜎𝐼𝑁𝑆 $/kWh 8 ✓ -
Mode for GC/IS PQ - ✓
𝑃𝑖𝐸𝑆,𝑚𝑎𝑥 MW 0.3 ✓ -
𝐸𝑆
𝑟 𝐸𝑆 - 5 ✓ ✓
𝐷𝑜𝐷𝑚𝑎𝑥 % 70 - ✓
𝜂𝑖𝐸𝑆 % 95 - ✓
𝑆𝑜𝐶 0 % 50 - ✓
𝜒 𝐸𝑆,𝑚𝑖𝑛 /𝜒 𝐸𝑆,𝑚𝑎𝑥 - -0.6/0.6 - ✓
171
TABLE G. 10: General parameters (continue)
Elements Parameters Unit/detail Values Level problem
outer inner
TR 𝑆 𝑇𝑅,𝑚𝑎𝑥 kVA 5300 - ✓
𝜒 𝑇𝑅,𝑚𝑖𝑛
/𝜒 𝑇𝑅,𝑚𝑎𝑥 - -0.6/0.6 - ✓
𝑃𝐹𝑖𝑇𝑅,1𝑠𝑡 - 0.75 - ✓
𝑃𝐹𝑖𝑇𝑅,4𝑡ℎ - 0.75 - ✓
𝑉𝑝𝑐𝑐 p.u. 1.05 - ✓
Network 𝑃𝑖𝑏𝑢𝑠,𝑚𝑎𝑥 kVA/ph 6.00a,4.00b ✓ -
𝐼𝑘=1→8 AΨ 500 - ✓
𝐼𝑘=9→32 AΨ 350 - ✓
SSW 𝑆𝑆𝑊
𝜎𝐼𝑁𝑉 $/SSW 64.3 ✓ -
𝑆𝑆𝑊
𝜎𝑂&𝑀 $/h 1.15 ✓ -
General 𝑁𝑇 h 24 - ✓
𝑇𝐿 year 10 ✓ -
𝐺𝑅𝐼𝐷
𝐾𝐶𝑂2 kg/kWh 143 ✓ -
𝜎 𝑐𝑢𝑟𝑡 $/kWh 3.4 - ✓
𝑃𝐶 𝑉𝐼 $/kV 1.1 - ✓
𝑆𝑏𝑎𝑠𝑒 kVA 1000 ✓ ✓
𝑉𝑏𝑎𝑠𝑒 kV 12.66 - ✓
𝑠𝑡
1 ,4
𝜑𝑠𝑒𝑙
𝑡ℎ
deg. 90 - ✓
𝑠𝑡
𝑛1𝑝𝑤,4
𝑡ℎ
- 15 - ✓
172

MR Y 38254 PDF

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

MR Y 38254 PDF

Transféré par

Droits d'auteur :

Formats disponibles

AC Microgrids Analysis, Optimization and Planning for

Doctorat en génie électrique

© Youthanalack Vilaisarn, 2022

Jérôme Cros, directeur de recherche

Table of Contents ............................................................................................................................ vii

List of Figures ................................................................................................................................... xi

List of Tables .................................................................................................................................. xiv

Remerciements ........................................................................................................................... xxxiv

Foreword ...................................................................................................................................... xxxv

Thesis Original Contribution .................................................................................................... xxxix

Chapter 1 Research Background and State of Art ........................................................................ 6

Chapter 2 Power Flow Algorithm for Islanded MGs .................................................................. 24

Chapter 3 Linear Power Flow Model for MG system ................................................................. 42

Chapter 4 MILP Model for MG’s Optimum Operation ............................................................. 50

Chapter 5 Resilience Oriented Microgrid Planning .................................................................... 78

Conclusion and Perspectives ........................................................................................................ 131

Bibliography .................................................................................................................................. 135

Appendix A: Element of square matrix ...................................................................................... 148

Appendix B: Wind, PV and load models..................................................................................... 150

Appendix C: Algorithm & Example for MGs Formation ......................................................... 151

Appendix D: Feasible operating regions of DDGs ..................................................................... 161

Appendix E: Linearization of absolute value operator .............................................................. 163

Appendix F: Decision variables for OPF/LOPF problem ......................................................... 164

ISW sub Isolate Switch at the upstream substation

Sets and Numbers

List of Symbols Used in Chapter 1

List of Symbols Used in Chapter 2

Variables and elements exclusive used for chapter 2:

List of Symbols Used in Chapter 3

List of Symbols Used in Chapter 4

Parameters and vector/matrix:

Function and Metric:

List of Symbols Used in Chapter 5

Parameters and vector/matrix:

Function and Metric:

Je tiens à remercier les membres de l’équipe du Programme Canadien de Bourse de la Francophonie

• Status: published with « doi.org/10.1016/j.ijepes.2020.106048 »

• Co-author: Morad Abdelaziz; contributions: supervised the development of the proposed

• Status: published with « doi:10.1049/oap-cired.2021.0019 »

• Co-author: Morad Abdelaziz; contributions: supervised the development of the proposed

• Status: published with « doi.org/10.1016/j.ijepes.2021.107674 »

• Co-author: Morad Abdelaziz; contributions: supervised the development of the proposed

• Status: Accepted for publication in IEEE Access by 30-July-2022

• Co-author: Morad Abdelaziz; contributions: supervised the development of the proposed

The general contribution of this thesis can be summarized as follows:

• Proposed a general framework using meta-heuristic multi-objective optimization for planning

0.1 Thesis motivation

0.2 Research Objective

• Objective 4: Proposing a general framework for resilience oriented microgrids planning in

A branch based power flow Objective 1:

Linear power flow model for Objective 2:

optimum operation for grid- Objective 3:

General planning framework for Objective 4:

FIGURE 0. 1: Basic outline of the research objectives

1.2 Passive and Active Distribution System in the Power System

FIGURE 1. 1: Different features of typical PDS(a) and ADS(b)

Upstream DG ESS Downstream

Distributed Distributed Energy Storage Sectionalizing Point of common

FIGURE 1. 2: General scheme of the basic microgrid

f/P and V/Q

𝑚𝑃 = (𝑓 𝑚𝑎𝑥 − 𝑓 𝑚𝑖𝑛 )⁄𝑃𝐺,𝑚𝑎𝑥 (1.3)

𝑛𝑞 = (𝑉 𝑚𝑎𝑥 − 𝑉 𝑚𝑖𝑛 )/𝑄 𝐺,𝑚𝑎𝑥 (1.4)

or by the inverse from,